commit c3902245002635b0e55093a5fb903bddcf46ece9
Author: Vladislav Kopnin <vlad.kopnen@mail.ru>
Date:   Sat Oct 24 01:17:36 2026 +0300

    Первичная версия: стилометрия реплик, LSA и протокол экспериментов

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..c835d73
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,16 @@
+.idea/
+.venv/
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.DS_Store
+
+lab1/nltk_data/
+lab1/data/processed/
+lab1/features/
+lab1/exports/
+lab1/results/*
+!lab1/results/README.md
+
+*.tar.gz
+*.tar.gz.sha256
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..6089314
--- /dev/null
+++ b/README.md
@@ -0,0 +1,56 @@
+# NLP: стилометрия реплик Stardew Valley
+
+Репозиторий лабораторной работы по автоматическому анализу текстов.
+Исследуется, можно ли стилометрически различить 34 персонажа Stardew Valley
+по отдельным репликам, если все реплики написаны одним автором.
+
+## Быстрый старт
+
+```bash
+cd lab1
+cat RUNBOOK.md
+```
+
+Пошаговая инструкция для запуска полной серии экспериментов:
+[`lab1/RUNBOOK.md`](lab1/RUNBOOK.md).
+
+Краткая установка:
+
+```bash
+cd lab1
+python -m venv .venv
+source .venv/bin/activate
+python -m pip install -r requirements.txt
+python scripts/download_nltk_data.py
+python scripts/check_environment.py
+python -m pytest -q
+bash scripts/run_all_experiments.sh
+```
+
+## Структура
+
+```text
+nlp/
+├── lab1/
+│   ├── stylometry/          # код: чтение, признаки, LSA, модели, метрики
+│   ├── tests/               # тесты
+│   ├── data/                # корпус и метаданные
+│   ├── configs/             # конфигурация
+│   ├── scripts/             # проверка окружения, запуск серии, упаковка
+│   ├── RUNBOOK.md           # инструкция для запуска
+│   ├── EXPERIMENTS.md       # протокол экспериментов
+│   ├── FEATURES.md          # описание признаков
+│   ├── METRICS_AND_CLASSIFIERS.md
+│   └── PRESENTATION_OUTLINE.md
+└── .venv/                   # локальное окружение, не коммитится
+```
+
+## Основные документы
+
+- [`lab1/RUNBOOK.md`](lab1/RUNBOOK.md) — как запускать всё с нуля.
+- [`lab1/EXPERIMENTS.md`](lab1/EXPERIMENTS.md) — протокол экспериментов.
+- [`lab1/METRICS_AND_CLASSIFIERS.md`](lab1/METRICS_AND_CLASSIFIERS.md) —
+  методическое описание метрик и классификаторов.
+- [`lab1/FEATURES.md`](lab1/FEATURES.md) — признаки и формулы.
+- [`lab1/PRESENTATION_OUTLINE.md`](lab1/PRESENTATION_OUTLINE.md) — структура
+  презентации.
diff --git a/lab1/.gitignore b/lab1/.gitignore
new file mode 100644
index 0000000..5c8e83e
--- /dev/null
+++ b/lab1/.gitignore
@@ -0,0 +1,11 @@
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.idea/
+nltk_data/
+data/processed/
+features/
+results/*
+!results/README.md
+
+exports/
diff --git a/lab1/EXPERIMENTS.md b/lab1/EXPERIMENTS.md
new file mode 100644
index 0000000..0481713
--- /dev/null
+++ b/lab1/EXPERIMENTS.md
@@ -0,0 +1,747 @@
+# Протокол экспериментов
+
+Документ описывает постановку, признаки, схему валидации, метрики и порядок
+запуска экспериментов. Он рассчитан на следующего исследователя, который будет
+проводить финальные серии опытов и готовить отчёт.
+
+Методические пояснения к метрикам и классификаторам вынесены в
+`METRICS_AND_CLASSIFIERS.md`. Здесь — практическая часть: что запускать, в
+каком порядке и как интерпретировать результаты.
+
+---
+
+## 1. Исследовательский вопрос
+
+**Вопрос.** Различимы ли 34 персонажа Stardew Valley по стилометрическим
+признакам отдельных реплик, если все реплики написаны одним автором?
+
+Это задача **классификации речевых масок персонажей**, а не авторской
+атрибуции. Автор один, но персонажи наделены разной лексикой, синтаксисом,
+междометиями, длиной реплик и пунктуацией. Положительный результат означает,
+что созданные автором речевые маски статистически различимы; он не означает,
+что у персонажей разные авторы.
+
+### Гипотезы
+
+| Обозначение | Гипотеза | Как проверяем |
+|---|---|---|
+| H1 | Функциональные слова, POS-признаки и символьные n-граммы различают персонажей выше случайного уровня | E2, E3, E4 |
+| H2 | На коротких репликах POS- и char-признаки устойчивее словарного богатства и читаемости | E3, E9 |
+| H3 | LSA по всем словам отражает тему, а LSA по POS и функциональным словам — стиль | E7 |
+| H4 | Более длинные реплики классифицируются лучше | E9 |
+| H5 | Ошибки группируются вокруг персонажей со сходным возрастом, полом, ролью или «человечностью» | E14 |
+
+Случайный уровень для 34 классов: доля правильных ответов около 2,9%,
+macro-F1 около 2,9%, top-3 около 8,8%. Любые результаты нужно сравнивать с
+этими значениями, а не с нулём.
+
+---
+
+## 2. Данные
+
+### 2.1. Текущий корпус
+
+| Параметр | Значение |
+|---|---|
+| Файл | `data/dialogues.csv` |
+| Кодировка | cp1251 |
+| Реплик | 2005 |
+| Персонажей | 34 |
+| Словоупотреблений | около 38,2 тыс. |
+| Средняя длина реплики | 19,1 слова |
+| Медиана | 17 слов |
+| Минимум / максимум | 3 / 78 слов |
+| Классы | 15–139 реплик |
+| Точные дубли реплик | 0 |
+
+Корпус уже прошёл ручную фильтрацию: удалены короткие и общие фразы. Точные
+правила очистки не сохранились, поэтому корпус используется как есть. Это
+ограничение нужно указать в отчёте: восстановить исходное распределение
+коротких реплик нельзя.
+
+Редкие классы: Mister Qi (15), Sandy (21), Dwarf (23), Wizard (24),
+Vincent (26), Krobus (28). При 5-блочной кросс-валидации у Mister Qi в каждом
+тестовом блоке около 3 реплик; метрики по нему будут иметь широкий
+доверительный интервал.
+
+### 2.2. Аудит
+
+```bash
+cd lab1
+python -m stylometry.cli audit
+```
+
+Результаты:
+
+- `data/processed/lines.csv` — нормализованные реплики;
+- `data/processed/audit.md` — отчёт с распределением классов, длин и сверкой
+  с `stats_per_character.xlsx`;
+- `data/processed/figures/class_distribution.png` — график.
+
+---
+
+## 3. Предобработка
+
+Корпус не перечищается. Выполняются только обратимые и не меняющие смысл
+операции:
+
+1. Приведение Unicode: многоточие, кавычки, тире, неразрывные пробелы.
+2. Замена плейсхолдеров `<PLAYER>`, `<FARM>`, `<PET>`, `<SPOUSE>`, `<TIME>`,
+   `<NOUN>`, `<NAME>` на слова `PLAYER`, `FARM` и т.д. Количество плейсхолдеров
+   сохраняется отдельным признаком.
+3. Схлопывание повторных пробелов.
+4. Токенизация NLTK: `word_tokenize`.
+5. Разбиение на предложения: `sent_tokenize`.
+6. Частеречная разметка: `pos_tag`; перевод тегов Penn Treebank в Universal
+   POS.
+7. Для POS-последовательностей подряд идущие знаки препинания схлопываются в
+   один тег `PUNCT`.
+
+Не выполняется:
+
+- удаление стоп-слов (они являются стилометрическим сигналом);
+- удаление пунктуации;
+- лемматизация по умолчанию (инфлексия — тоже стиль);
+- удаление коротких реплик;
+- приведение регистра к нижнему для всех признаков.
+
+### 3.1. Ресурсы NLTK
+
+```bash
+cd lab1
+python scripts/download_nltk_data.py
+```
+
+Скачиваются `punkt`, `punkt_tab`, `averaged_perceptron_tagger`,
+`averaged_perceptron_tagger_eng`, `stopwords` в `lab1/nltk_data`. Если ресурсы
+недоступны, токенизатор переходит в резервный режим на регулярных выражениях,
+а POS-признаки отключаются. Для финальных экспериментов нужен NLTK-режим:
+проверьте, что в метаданных запуска `pos_available=true`.
+
+---
+
+## 4. Признаковое пространство
+
+### 4.1. Ручные стилометрические признаки
+
+Всего около 454 признаков. Они не обучаются на данных, поэтому могут быть
+посчитаны заранее.
+
+| Группа | Примеры | Статус |
+|---|---|---|
+| Длина | число токенов, слов, символов, предложений; средняя длина предложения; доля длинных слов | Основная, но чувствительна к длине |
+| Лексика | распределение длин слов, доля слов с апострофом, заглавных, цифр | Основная |
+| Словарное богатство | TTR, Guiraud, Herdan, Yule's K, Simpson's D, hapax/dis, Honoré | Предварительный |
+| Символы | частоты букв a–z, доля заглавных, цифр, пунктуации, апострофов | Основная |
+| Пунктуация | запятые, точки, восклицательные, вопросительные, многоточия, повторная пунктуация | Основная |
+| Функциональные слова | около 150 закрытых слов и 12 категорий | Основная |
+| Местоимения | 1-е, 2-е, 3-е лицо, возвратные, притяжательные, отношение I/you | Основная |
+| Части речи | Universal POS, Penn Treebank, модальность, время, герундий, причастие | Основная |
+| Синтаксис-прокси | число клауз, подчинительные союзы, вопрос/восклицание, пассив-прокси | Основная |
+| Читаемость | Flesch, FK grade, Gunning Fog, SMOG, Coleman-Liau, ARI | Предварительный |
+| Разговорные маркеры | плейсхолдеры, ALL CAPS, многоточия, междометия, смех, дискурсивные маркеры, хеджи, интенсификаторы, вежливость, отрицания, сокращения | Основная |
+
+Признаки `vocabulary` и `readability` помечены как предварительные: на репликах
+длиной 10–20 слов они нестабильны. Их влияние нужно проверять отдельно, а не
+включать в основной вывод без оговорок.
+
+### 4.2. Символьные n-граммы
+
+TF-IDF по `char_wb`, диапазон 2–5, `min_df=3`, до 5000 признаков, L2-норма.
+Хорошо отражают морфологию, пунктуацию и орфографические привычки.
+
+### 4.3. POS n-граммы
+
+TF-IDF по последовательностям Universal POS, диапазон 1–3, `min_df=3`, до 500
+признаков, L2-норма. Это относительно устойчивый к теме синтаксический сигнал.
+
+### 4.4. Словарь признаков
+
+```bash
+python -m stylometry.cli features
+```
+
+Создаёт:
+
+- `features/stylometric.csv` — матрица 2005 × 454;
+- `features/feature_dictionary.csv` — название, группа, описание, статус.
+
+Словарь нужно использовать в отчёте: для каждой группы указать, какие
+признаки и почему выбраны.
+
+---
+
+## 5. Латентно-семантический анализ
+
+LSA реализован как:
+
+```text
+TF-IDF -> TruncatedSVD(k) -> L2-нормализация
+```
+
+Варианты:
+
+| Вариант | Колонка | n-граммы | Что отражает |
+|---|---|---|---|
+| `word` | `text` | 1 | тему и стиль |
+| `word2` | `text` | 1–2 | тему и устойчивые сочетания |
+| `content` | `content_text` | 1–2 | содержание после удаления стоп-слов |
+| `fw` | `fw_text` | 1–2 | синтаксис через функциональные слова |
+| `char` | `text` | 2–5 (символы) | морфологию и пунктуацию |
+| `pos` | `pos_text` | 1–4 (POS) | синтаксический стиль |
+
+Основные параметры: `sublinear_tf=true`, `min_df=3`, `max_df=0.9`,
+`max_features=2000–8000`, `n_components=50–300`. Все параметры настраиваются
+только на обучающих блоках; для интерпретационных графиков можно обучить LSA
+на всём корпусе, но это нельзя использовать для оценки качества.
+
+### 5.1. Что проверять в LSA
+
+1. **График объяснённой дисперсии и накопленная дисперсия.** Сколько
+   компонент объясняют основную часть дисперсии.
+2. **Top-термины компонент.** Интерпретируемость осей: тема, эмоциональность,
+   формальность, длина.
+3. **Внешняя оценка.** Macro-F1 классификатора при разных `k`.
+4. **Сравнение вариантов.** Если `lsa_word` и `lsa_content` значительно лучше
+   `lsa_pos` и `lsa_fw`, классификация в существенной степени опирается на
+   тему, а не на стиль.
+5. **Стабильность.** Bootstrap по репликам и углы между подпространствами
+   (опционально; при малом числе компонент полезно).
+
+### 5.2. Команды
+
+```bash
+python -m stylometry.cli lsa --variant word --components 100
+python -m stylometry.cli lsa --variant pos  --components 50
+python -m stylometry.cli lsa --variant char --components 50
+python -m stylometry.cli lsa-sweep --variants word,content,fw,char,pos \
+    --components 50,100,200 --model logreg --repeats 1
+```
+
+Результаты: матрицы `features/lsa_<variant>_k<k>.csv`, наиболее
+информативные термины компонент, объяснённая дисперсия, график объяснённой
+дисперсии.
+
+---
+
+## 6. Классификаторы
+
+В основном сравнении участвуют:
+
+| Модель | Краткая характеристика | Вероятности |
+|---|---|---|
+| `dummy` | предсказывает самый частый класс | да |
+| `multinomial_nb` | наивный байесовский классификатор для неотрицательных признаков | да |
+| `logreg` | логистическая регрессия с L2 и балансировкой классов | да |
+| `linear_svc` | линейный метод опорных векторов | нет, есть решающая функция |
+| `random_forest` | случайный лес | да |
+| `extra_trees` | сверхслучайные деревья | да |
+| `knn` | k ближайших соседей с косинусным расстоянием | да |
+
+Для MultinomialNB автоматически исключаются LSA-блоки (компоненты могут быть
+отрицательными) и используется MinMax-масштабирование. Для остальных моделей
+ручные признаки стандартизуются и L2-нормируются, а TF-IDF- и LSA-блоки уже
+нормированы. Это нужно, чтобы блоки разной размерности не подавляли друг
+друга.
+
+Подробное описание моделей, их формул, гиперпараметров и ограничений — в
+`METRICS_AND_CLASSIFIERS.md`.
+
+### 6.1. Сетки гиперпараметров
+
+| Модель | Параметры |
+|---|---|
+| `logreg` | `C ∈ {0.1, 1, 10}` |
+| `linear_svc` | `C ∈ {0.1, 1, 10}` |
+| `random_forest`, `extra_trees` | `max_depth ∈ {None, 30}`, `min_samples_leaf ∈ {1, 2}` |
+| `multinomial_nb` | `alpha ∈ {0.1, 1, 3}` |
+| `knn` | `n_neighbors ∈ {3, 5, 9}` |
+
+Настройка выполняется внутри вложенной кросс-валидации. Для одиночного запуска
+`run` без `--nested` гиперпараметры фиксированы значениями по умолчанию; это
+быстрее, но для финального сравнения нужен `--nested`.
+
+---
+
+## 7. Схема валидации
+
+### 7.1. Основная схема
+
+```python
+RepeatedStratifiedKFold(n_splits=5, n_repeats=5, random_state=42)
+```
+
+Почему 5 блоков:
+
+- минимальный класс — 15 реплик;
+- в каждом тестовом блоке около 3 реплик на класс;
+- все 34 класса присутствуют и в обучении, и в тесте;
+- 10 блоков дали бы 1–2 реплики на класс и ещё более шумные оценки.
+
+Почему повторная:
+
+- один прогон 5-блочной CV даёт одну оценку; повторность позволяет оценить
+  разброс;
+- для финального отчёта используйте 5 повторов; для быстрых проверок — 1–3.
+
+### 7.2. Вложенная кросс-валидация
+
+```python
+outer = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
+inner = StratifiedKFold(n_splits=3, shuffle=True, random_state=42)
+```
+
+Внешние блоки — оценка качества; внутренние — настройка гиперпараметров.
+Такой протокол уменьшает оптимистическое смещение, которое возникает, если
+настраивать гиперпараметры на тех же данных, где измеряется качество.
+
+### 7.3. Контроль утечек
+
+- TF-IDF, SVD, масштабирование и отбор признаков обучаются только на
+  обучающем блоке.
+- Ручные признаки и POS-теги не используют метки; их можно считать заранее.
+- Точных дублей реплик в корпусе нет, поэтому одна и та же реплика не может
+  попасть в обучение и тест.
+- При появлении почти одинаковых реплик нужно проверить `StratifiedGroupKFold`
+  с группами по нормализованному тексту. В текущем корпусе точных дублей нет.
+- Нельзя выбирать признаки или `k` LSA по тестовым метрикам. Все решения — по
+  внутренней кросс-валидации.
+
+---
+
+## 8. Метрики
+
+Подробные определения, формулы и интерпретация — в
+`METRICS_AND_CLASSIFIERS.md`. Кратко:
+
+| Метрика | Роль |
+|---|---|
+| Macro-F1 | Основная метрика: каждый из 34 классов равноправен |
+| Сбалансированная доля правильных ответов | Средняя полнота по классам; удобна для редких персонажей |
+| Доля правильных ответов | Нужна для сравнения с литературой |
+| MCC | Устойчив к дисбалансу, учитывает все клетки матрицы ошибок |
+| Cohen's kappa | Согласие с поправкой на случай |
+| Top-3, top-5 | Важны при 34 классах |
+| Log loss, Brier | Качество вероятностей и калибровка |
+| Точность, полнота и F-мера по классам | Показывают, какие персонажи проваливаются |
+| Матрица ошибок | Анализ пар перепутывания |
+
+Основной отчётный результат — **pooled out-of-fold macro-F1** с 95%
+бутстрэп-интервалом и среднее ± стандартное отклонение по повторам. Среднее
+macro-F1 по отдельным блокам тоже полезно, но на класс приходится около трёх
+объектов, поэтому оно шумное.
+
+### 8.1. Статистические сравнения
+
+- **Мак-Немар** — парное сравнение двух моделей на одних и тех же тестовых
+  репликах.
+- **Уилкоксон** — парное сравнение результатов по блокам; описательная
+  проверка стабильности.
+- **Фридман** — сравнение трёх и более моделей.
+- **Поправка Холма—Бонферрони** — при множественных сравнениях.
+- **Бутстрэп-интервалы** — основной способ показать неопределённость.
+
+Не сравнивайте модели только по доле правильных ответов: при дисбалансе и
+34 классах она может вводить в заблуждение. Пример: модель, предсказывающая
+только Abigail, имеет долю правильных ответов около 7%, но macro-F1 около
+0,004.
+
+---
+
+## 9. Матрица экспериментов
+
+Все команды выполняются из `lab1`. Перед серией полезно один раз выполнить
+`audit` и `features`.
+
+### E0. Аудит корпуса
+
+```bash
+python -m stylometry.cli audit
+```
+
+Что проверить: число реплик и классов, распределение длин, отсутствие дублей,
+совпадение с `stats_per_character.xlsx`.
+
+### E1. Базовые уровни
+
+```bash
+python -m stylometry.cli run --preset length --model logreg --repeats 5
+python -m stylometry.cli run --preset char   --model multinomial_nb --repeats 3
+python -m stylometry.cli run --preset pos    --model multinomial_nb --repeats 3
+python -m stylometry.cli run --preset fw     --model logreg --repeats 3
+python -m stylometry.cli run --preset length --model dummy --repeats 1
+```
+
+Смысл: понять, сколько сигнала даёт только длина, только символьные n-граммы,
+только POS, только функциональные слова и случайная/частотная базовая модель.
+
+### E2. Классическая стилометрия
+
+```bash
+python -m stylometry.cli run --preset stylometry --model logreg --repeats 5
+python -m stylometry.cli run --preset stylometry --model linear_svc --repeats 5
+```
+
+Проверяет H1 на ручных признаках без n-грамм и LSA.
+
+### E3. Поочерёдное исключение групп признаков
+
+```bash
+python -m stylometry.cli ablate --preset stylometry_pos_lsa --model logreg --repeats 3
+```
+
+Для каждой группы признаков строится вариант «без этой группы». Результат —
+`results/ablation_*/ablation.csv` с колонкой `delta_macro_f1`. Отрицательное
+значение означает, что удаление группы ухудшило качество; положительное — что
+группа мешала (возможно, из-за шума или переобучения).
+
+### E4. POS-признаки
+
+Сравните:
+
+```bash
+python -m stylometry.cli run --preset stylometry     --model logreg --repeats 5
+python -m stylometry.cli run --preset stylometry_pos --model logreg --repeats 5
+```
+
+Разница macro-F1 показывает вклад символьных и POS n-грамм.
+
+### E5. LSA: варианты и число компонент
+
+```bash
+python -m stylometry.cli lsa-sweep \
+    --variants word,content,fw,char,pos \
+    --components 50,100,200 \
+    --model logreg --repeats 1
+```
+
+Затем лучшие `k` и варианты проверить с 5 повторами через `run`:
+
+```bash
+python -m stylometry.cli run --preset lsa_word --model logreg --repeats 5
+```
+
+Для `lsa_word` и `lsa_pos` число компонент задаётся в `FeatureConfig`; в
+командной строке текущая версия использует значение из `presets`
+(`lsa_components=100`). Если нужно другое `k`, используйте `lsa-sweep` или
+измените `FeatureConfig` в сценарии.
+
+### E6. Сравнение классификаторов
+
+```bash
+python -m stylometry.cli benchmark --preset stylometry_pos_lsa \
+    --models logreg,linear_svc,random_forest,multinomial_nb,knn \
+    --repeats 3
+```
+
+Результат: `results/benchmark_*/summary.csv`, отдельные каталоги по моделям.
+
+### E7. Стиль или тема
+
+Сравните LSA-варианты:
+
+```bash
+python -m stylometry.cli lsa-sweep \
+    --variants word,content,fw,char,pos \
+    --components 50,100,200 --model logreg --repeats 1
+```
+
+Ожидаемая интерпретация:
+
+- `word` и `content` лучше — вклад темы велик;
+- `pos` и `fw` сопоставимы — стиль/синтаксис достаточно информативны;
+- `char` полезен для морфологии и пунктуации.
+
+### E8. Объединение блоков
+
+```bash
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --repeats 5
+python -m stylometry.cli run --preset all --model logreg --repeats 3
+```
+
+Сравните с E2 и E5. Если объединение не лучше лучшего отдельного блока,
+проверьте веса блоков и размерности.
+
+### E9. Влияние длины
+
+1. Запустите `run --preset length`, `--preset stylometry`,
+   `--preset stylometry_pos_lsa`.
+2. Для сохранённого запуска выполните:
+
+```bash
+python -m stylometry.cli length-report --run-dir results/<каталог_запуска>
+```
+
+Команда сама соединяет предсказания с `data/processed/lines.csv`, разбивает
+реплики на группы 0–5, 6–10, 11–15, 16–20, 21–30, 31+ слов и считает
+macro-F1 с бутстрэп-интервалом. В каждой группе macro-F1 считается по
+классам, которые в ней присутствуют; поэтому сравнение групп с разным
+набором персонажей нужно интерпретировать осторожно. Результат сохраняется
+в `length_bins.csv` внутри каталога запуска.
+
+Ожидание: качество может зависеть от длины, но монотонность не гарантирована.
+В предварительном прогоне объединённого набора лучшие значения оказались
+в группах 21–30 и 31+ слов, а группа 0–5 слов дала широкий интервал из-за
+малого числа объектов. Если длина вносит существенный вклад, сравните
+результаты с `--preset length` и без length-признаков (поочерёдное исключение
+групп).
+
+### E10. Дисбаланс классов
+
+```bash
+python -m stylometry.cli benchmark --preset stylometry_pos_lsa \
+    --models logreg,linear_svc,random_forest --repeats 3
+```
+
+Смотрите не только macro-F1, но и F-меру по классам, сбалансированную долю
+правильных ответов и support. Сравните `class_weight="balanced"` (в коде
+включено для logreg, SVM, RF) с
+вариантом без балансировки: для этого нужно изменить `make_classifier` или
+передать параметр.
+
+### E11. Дельта Барроуза
+
+```bash
+python -m stylometry.cli delta --features 100,200,500 --metrics manhattan,cosine --repeats 3
+```
+
+Классическая стилометрическая базовая модель. Ожидается, что на коротких репликах
+он уступает ML-моделям; важен как независимая проверка.
+
+### E12. Верификация «свой/чужой»
+
+```bash
+python -m stylometry.cli verify --preset stylometry_pos_lsa
+```
+
+Для каждой тестовой реплики считается близость к центроидам всех персонажей.
+Положительный класс — центроид своего персонажа, отрицательный — чужие.
+Метрики: ROC-AUC, EER, TAR@FAR=1%, top-1. Это более мягкая
+постановка, чем 34-классовая классификация.
+
+### E13. Кластеризация
+
+```bash
+python -m stylometry.cli cluster --preset stylometry_pos_lsa
+```
+
+Строятся центроиды персонажей, иерархическая кластеризация и k-means.
+Метрики: ARI, NMI, силуэт. Кластеризация носит предварительный характер: она показывает, какие
+персонажи близки в признаковом пространстве, но не заменяет классификацию.
+
+### E14. Анализ ошибок
+
+После любого запуска откройте:
+
+- `per_class_report.csv` — precision, recall, F1, support по персонажам;
+- `confusion_matrix_normalized.csv` — доли перепутывания;
+- `top_confusions.csv` — самые частые пары;
+- `figures/confusion_matrix.png` — тепловая карта.
+
+Что искать:
+
+- персонажи с малым support и низким F1;
+- пары с высокой взаимной ошибкой (например, дети, нелюди, персонажи с
+  похожей ролью);
+- влияние длины: короткие реплики чаще путаются;
+- признаки, важные для лучшей модели (коэффициенты логистической регрессии,
+  permutation importance для деревьев).
+
+Для анализа по метаданным (пол, возраст, вид) используйте:
+
+```bash
+python -m stylometry.cli subgroup-report \
+    --run-dir results/<каталог_запуска> --group gender
+python -m stylometry.cli subgroup-report \
+    --run-dir results/<каталог_запуска> --group age_group
+python -m stylometry.cli subgroup-report \
+    --run-dir results/<каталог_запуска> --group species
+```
+
+Команда сохраняет `subgroup_<group>_metrics.csv` и
+`subgroup_<group>_confusions.csv`. Метаданные лежат в
+`data/metadata/characters.csv`; это экспертные пометки, а не часть корпуса,
+поэтому они не используются как признаки классификатора.
+
+### E15. Статистическое сравнение
+
+Для двух лучших моделей:
+
+```bash
+python -m stylometry.cli compare \
+    --run-a results/<каталог_запуска_A> \
+    --run-b results/<каталог_запуска_B>
+```
+
+Команда соединяет out-of-fold предсказания по `line_id`, проверяет
+совпадение истинных классов и считает точный критерий Мак-Немара и
+бутстрэп-интервал разницы macro-F1. Результат сохраняется в
+`comparison_with_other.json`. Для корректного парного сравнения запускайте
+модели с одинаковыми `--folds` и `--seed`.
+
+Для трёх и более моделей используйте `friedman_test` из
+`stylometry/stats.py` по метрикам блоков и поправку Холма.
+
+### E16. Вложенная кросс-валидация
+
+```bash
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --nested
+python -m stylometry.cli run --preset stylometry_pos_lsa --model linear_svc --nested
+```
+
+Это финальная проверка с настройкой гиперпараметров. Сохраняются
+`best_params.json` и метрики внешних блоков.
+
+### E17. Дополнительно: агрегация в псевдодокументы
+
+Основной протокол — уровень реплик. Если понадобится формально закрыть
+требование «не менее 30 предложений на текст», можно реализовать отдельный
+сценарий: склеить реплики одного персонажа в псевдодокументы по 30 предложений
+и повторить E2–E9. Этот сценарий **не включён** в основной код, потому что
+меняет единицу анализа и уменьшает число объектов. Его результаты следует
+рассматривать как robustness check, а не как замену основного эксперимента.
+
+---
+
+## 10. Контроль утечек и воспроизводимости
+
+Перед финальным запуском проверьте:
+
+- [ ] `data/processed/tokenized.pkl` соответствует текущему SHA-256 корпуса.
+- [ ] В метаданных запуска `pos_available=true`; POS-признаки не отключились.
+- [ ] TF-IDF, SVD и масштабирование обучаются внутри блоков (это обеспечивает
+      `Pipeline`; не выносите их наружу).
+- [ ] `k` LSA и гиперпараметры выбраны по внутренней CV, а не по тесту.
+- [ ] В корпусе нет точных дублей; при появлении почти одинаковых реплик
+      используется группировка.
+- [ ] Для финальных чисел выполнено не менее 3–5 повторов.
+- [ ] Сохранены `run_metadata.json`, `fold_metrics.csv`, `pooled_metrics.csv`,
+      `oof_predictions.csv`, `confusion_matrix.csv`.
+- [ ] Указаны версии библиотек и seed.
+
+---
+
+## 11. Текущие технические прогоны
+
+Ниже — предварительные результаты одного повтора 5-блочной CV. Они нужны
+только для проверки работоспособности конвейера и **не являются финальными**:
+один повтор даёт широкий интервал, а гиперпараметры не настраивались.
+
+| Набор признаков | Модель | Macro-F1 | Доля правильных ответов | Top-3 |
+|---|---|---:|---:|---:|
+| `stylometry` | logreg | 0,1002 | 0,0988 | 0,2389 |
+| `stylometry_pos` | logreg | 0,1780 | 0,1681 | 0,3367 |
+| `stylometry_pos_lsa` | logreg | 0,1802 | 0,1706 | 0,3387 |
+| `lsa_word`, k=50 | logreg | 0,0618 | 0,0628 | 0,1835 |
+| `lsa_pos`, k=50 | logreg | 0,0443 | 0,0459 | 0,1202 |
+| `char` | multinomial_nb | 0,0280 | 0,0973 | 0,2464 |
+
+Наблюдения:
+
+- символьные и POS n-граммы дают основной прирост относительно ручных
+  признаков;
+- LSA по словам и POS сама по себе слабее, но в объединении даёт небольшой
+  прирост;
+- доля правильных ответов у `char + multinomial_nb` около 9,7%, но
+  сбалансированная доля правильных ответов 4,7% и macro-F1 2,8%: модель
+  предсказывает частые классы. Это иллюстрация того, почему долю правильных
+  ответов нельзя использовать как единственную метрику.
+
+Верификация для `stylometry` (5 блоков): ROC-AUC ≈ 0,648, EER ≈ 0,395,
+top-1 ≈ 0,093. Это выше случайного уровня (ROC-AUC 0,5), но далеко от
+практически надёжного различения.
+
+Поочерёдное исключение групп для `stylometry` (один повтор): удаление группы `characters`
+снижает macro-F1 примерно на 0,015, удаление `vocabulary` — примерно на
+0,008; остальные изменения в пределах шума. Эти числа нужно перепроверить с
+несколькими повторами.
+
+---
+
+## 12. Как интерпретировать результаты
+
+1. **Сравнивайте с случайным уровнем.** Для 34 классов доля правильных
+   ответов 2,9%, macro-F1 2,9%, top-3 8,8%.
+2. **Основной вывод — по macro-F1 и доверительному интервалу**, а не по
+   доле правильных ответов.
+3. **Смотрите F-меру по классам.** Если редкие персонажи не различаются, это
+   важный вывод, а не техническая деталь.
+4. **Сравнивайте стиль и тему.** Если `lsa_word`/`lsa_content` значительно
+   лучше `lsa_pos`/`lsa_fw`, классификатор опирается на содержание.
+5. **Проверяйте длину.** Если `length` даёт заметный сигнал, часть результата
+   объясняется длиной реплик, а не синтаксисом.
+6. **Не переинтерпретируйте небольшие различия.** При 2005 объектах и 34
+   классах разница в 1–2 процентных пункта может быть шумом; нужны
+   доверительные интервалы и парные тесты.
+7. **Помните про одного автора.** Результат говорит о различимости персонажей,
+   а не о разных авторах.
+
+---
+
+## 13. Ограничения
+
+- Короткие тексты: признаки словарного богатства и читаемости нестабильны.
+- Один автор: это классификация речевых масок, не авторская атрибуция.
+- Ручная фильтрация корпуса: короткие и общие фразы удалены, правила не
+  восстановлены.
+- Дисбаланс классов: 15–139 реплик; редкие классы дают широкие интервалы.
+- Нет идентификаторов сцен: нельзя учесть корреляцию реплик внутри одной
+  сцены.
+- NLTK-теггер ошибается на разговорных конструкциях.
+- LSA может отражать тему; сравнение вариантов обязательно.
+- Кластеризация и t-SNE — предварительный анализ, не доказательство.
+
+---
+
+## 14. Чек-лист перед защитой
+
+- [ ] В отчёте описаны корпус, ручная фильтрация и её ограничения.
+- [ ] Указано, что задача — character idiolect, а не авторская атрибуция.
+- [ ] Перечислены группы признаков и обоснован выбор.
+- [ ] Описаны варианты LSA и способ выбора числа компонент.
+- [ ] Указаны классификаторы, гиперпараметры и схема валидации.
+- [ ] Основная метрика — macro-F1; приведены доверительные интервалы.
+- [ ] Есть отчёт по классам и матрица ошибок.
+- [ ] Есть сравнение стиля и темы (`lsa_word` vs `lsa_pos`/`lsa_fw`).
+- [ ] Есть анализ длины и поочерёдное исключение групп признаков.
+- [ ] Есть статистическое сравнение моделей.
+- [ ] Сформулированы ограничения и направления улучшения.
+- [ ] Код, отчёт и презентация воспроизводимы по `README.md`.
+
+---
+
+## 15. Куда сохраняются результаты
+
+```text
+results/
+├── <preset>_<model>_<timestamp>/
+│   ├── run_metadata.json
+│   ├── fold_metrics.csv
+│   ├── fold_summary.csv
+│   ├── pooled_metrics.csv
+│   ├── bootstrap_ci.json
+│   ├── oof_predictions.csv
+│   ├── per_class_report.csv
+│   ├── confusion_matrix.csv
+│   ├── confusion_matrix_normalized.csv
+│   ├── top_confusions.csv
+│   ├── summary.md
+│   └── figures/
+├── benchmark_<preset>_<timestamp>/
+│   ├── summary.csv
+│   └── <model>/
+├── ablation_<preset>_<timestamp>/
+│   └── ablation.csv
+├── lsa_sweep_<timestamp>/
+│   └── lsa_sweep.csv
+├── delta_<timestamp>/
+│   └── delta_sweep.csv
+├── verification_<preset>_<timestamp>/
+│   ├── fold_metrics.csv
+│   ├── pair_scores.csv
+│   └── summary.json
+└── clustering_<preset>_<timestamp>/
+    ├── clustering_metrics.csv
+    ├── character_centroids.csv
+    └── figures/
+```
diff --git a/lab1/FEATURES.md b/lab1/FEATURES.md
new file mode 100644
index 0000000..55e58da
--- /dev/null
+++ b/lab1/FEATURES.md
@@ -0,0 +1,294 @@
+# Стилометрические признаки: состав и обоснование
+
+Документ дополняет `EXPERIMENTS.md` и `METRICS_AND_CLASSIFIERS.md`. Здесь
+описано, какие признаки вычисляются, по каким формулам и почему они выбраны.
+Полный машинно-читаемый список с именами и описаниями создаётся командой:
+
+```bash
+python -m stylometry.cli features
+# результат: features/feature_dictionary.csv
+```
+
+Все ручные признаки вычисляются по одной реплике. Они не используют метки
+классов, поэтому могут быть посчитаны заранее, до разбиения на обучающую и
+тестовую выборки.
+
+---
+
+## 1. Признаки длины
+
+| Признак | Формула / смысл |
+|---|---|
+| `len_tokens` | Число токенов, включая пунктуацию |
+| `len_words` | Число токенов, содержащих букву или цифру |
+| `len_chars` | Длина строки в символах |
+| `len_sentences` | Число предложений |
+| `len_mean_sentence_words` | Средняя длина предложения в словах |
+| `len_std_sentence_words` | Стандартное отклонение длины предложения |
+| `len_min/max/median_sentence_words` | Минимум, максимум, медиана |
+| `len_pct_sentences_lt5` | Доля предложений короче 5 слов |
+| `len_pct_sentences_gt20` | Доля предложений длиннее 20 слов |
+| `len_mean/std/median_word_chars` | Длина слова в символах |
+| `len_pct_words_ge7/10/13` | Доля слов длиной не менее 7/10/13 символов |
+| `len_pct_single_letter_words` | Доля односимвольных слов |
+
+**Зачем.** Длина реплики — сильный, но не всегда желательный признак. Она
+различает детей и взрослых, но может стать прокси персонажа, а не стиля.
+Поэтому в протоколе предусмотрены ablation без length-признаков и анализ по
+группам длины.
+
+---
+
+## 2. Лексические признаки
+
+| Признак | Смысл |
+|---|---|
+| `lex_word_length_1` … `lex_word_length_10plus` | Доли слов соответствующей длины |
+| `lex_pct_apostrophe_words` | Доля слов с апострофом (сокращения, притяжательные) |
+| `lex_pct_capitalized_words` | Доля слов с заглавной буквы |
+| `lex_pct_all_caps_words` | Доля слов в верхнем регистре |
+| `lex_pct_hyphenated_words` | Доля слов с дефисом |
+| `lex_pct_digit_words` | Доля слов с цифрами |
+
+**Зачем.** Распределение длин слов и орфографические привычки устойчивы к теме
+и хорошо дополняют частотные признаки.
+
+---
+
+## 3. Словарное богатство
+
+Обозначения: \(N\) — число словоупотреблений, \(V\) — число типов,
+\(V_1\) — число гапаксов (слов, встретившихся один раз), \(V_2\) — число
+дис-гапаксов.
+
+| Признак | Формула |
+|---|---|
+| `vocab_ttr` | \(V/N\) |
+| `vocab_guiraud` | \(V/\sqrt{N}\) |
+| `vocab_herdan` | \(\log V / \log N\) |
+| `vocab_yule_k` | \(10^4(\sum_i i^2 V_i - N)/N^2\) |
+| `vocab_simpson_d` | \(\sum_i V_i(V_i-1)/(N(N-1))\) |
+| `vocab_hapax_ratio` | \(V_1/V\) |
+| `vocab_dis_ratio` | \(V_2/V\) |
+| `vocab_honore_r` | \(100\log N/(1 - V_1/V)\) |
+| `vocab_top10_coverage` | Доля токенов, покрытая 10 самыми частыми типами |
+| `vocab_mean_word_frequency` | \(N/V\) |
+
+**Зачем.** Классические меры богатства словаря. **Ограничение:** на репликах
+длиной 10–20 слов они нестабильны и сильно зависят от длины. Поэтому они
+помечены как предварительные и проверяются отдельно.
+
+---
+
+## 4. Символьные признаки
+
+| Признак | Смысл |
+|---|---|
+| `char_freq_a` … `char_freq_z` | Относительные частоты латинских букв |
+| `char_upper_rate` | Доля заглавных среди букв |
+| `char_digit_rate` | Доля цифр среди символов |
+| `char_punctuation_rate` | Доля символов пунктуации |
+| `char_apostrophe_rate` | Доля апострофов |
+| `char_quote_rate` | Доля кавычек |
+
+**Зачем.** Частоты букв и оформление слабо зависят от темы и хорошо
+дополняют словарные признаки. Символьные n-граммы (2–5) вынесены в отдельный
+TF-IDF-блок.
+
+---
+
+## 5. Пунктуация
+
+Считаются частоты на предложение и на токен:
+
+- запятая, точка, восклицательный и вопросительный знаки;
+- двоеточие и точка с запятой;
+- многоточие (последовательность из трёх и более точек);
+- апостроф, кавычки, дефис, скобки;
+- повторная пунктуация (`!!`, `??`, `?!`);
+- доля предложений, заканчивающихся вопросом или восклицанием.
+
+**Зачем.** Пунктуация — один из самых устойчивых авторских/персонажных
+маркеров. В игровых репликах она особенно важна: многоточия, восклицания и
+повторы часто отражают эмоциональность персонажа.
+
+---
+
+## 6. Функциональные слова
+
+Используется фиксированный список закрытых классов:
+
+- артикли, детерминативы, предлоги, союзы;
+- личные, притяжательные и возвратные местоимения;
+- вспомогательные глаголы и модальные;
+- отрицания и сокращения;
+- вопросительные слова и дискурсивные маркеры.
+
+Для каждого слова считается относительная частота \(c_w/N\). Дополнительно
+считаются частоты категорий.
+
+**Зачем.** Это ядро классической стилометрии: закрытые классы слов мало
+зависят от темы, но отражают синтаксические привычки. Метод дельты Барроуза
+также строится на частотных словах.
+
+---
+
+## 7. Местоимения
+
+| Признак | Смысл |
+|---|---|
+| `pron_first_singular_rate` | Я, меня, мой, myself |
+| `pron_first_plural_rate` | Мы, нас, наш |
+| `pron_second_rate` | Ты, вас, ваш |
+| `pron_third_singular_rate` | Он, она, оно |
+| `pron_third_plural_rate` | Они |
+| `pron_possessive_rate` | Притяжательные |
+| `pron_reflexive_rate` | Возвратные |
+| `pron_i_you_log_ratio` | \(\log((n_I+1)/(n_{you}+1))\) |
+| `pron_first_person_rate` | Все формы первого лица |
+| `pron_third_person_rate` | Все формы третьего лица |
+| `pron_wh_rate` | Вопросительные слова |
+
+**Зачем.** Соотношение «я — ты — он» отражает ориентацию речи: на себя, на
+собеседника или на третьих лиц. Это содержательно интерпретируемый признак.
+
+---
+
+## 8. Части речи
+
+Используются две системы:
+
+- **Universal POS** (16 тегов): ADJ, ADP, ADV, AUX, CONJ, DET, INTJ, NOUN,
+  NUM, PRON, PRT, PUNCT, SCONJ, SYM, VERB, X;
+- **Penn Treebank** (36 тегов): NN, NNS, NNP, VB, VBD, VBG, VBN, VBP, VBZ,
+  JJ, RB, PRP, IN, DT, CC, MD и другие.
+
+Считаются:
+
+- доли каждого UPOS- и PTB-тега;
+- отношения «существительное/глагол», «прилагательное/наречие»;
+- модальность, вспомогательность, прошедшее/настоящее время;
+- герундий, причастие, сравнительная и превосходная степень;
+- доля местоимений и содержательных слов.
+
+**Зачем.** Частеречные профили — стандартный стилометрический признак. Они
+устойчивее к теме, чем словарные частоты, и хорошо работают на коротких
+текстах. Дополнительно строятся POS n-граммы (1–3) и POS-LSA.
+
+---
+
+## 9. Синтаксические признаки-прокси
+
+Полный синтаксический анализ не используется. Вместо него считаются:
+
+- число клауз на предложение (по сочинительным и подчинительным союзам);
+- число подчинительных и сочинительных союзов на предложение;
+- доля предложений, начинающихся с союза;
+- доля вопросов и восклицаний;
+- модальность и отрицание;
+- частоты `that`, `to`, окончаний `-ing`, `-ed`, `-ly`;
+- пассив-прокси: форма `be` + причастие;
+- вопросительные хвосты (`right?`, `isn't it?`);
+- императив-прокси: предложение начинается с глагола из списка и не имеет
+  подлежащего-местоимения;
+- средняя и максимальная длина предложения.
+
+**Зачем.** Эти признаки приближают синтаксическую сложность без тяжёлых
+зависимостей. Они менее точны, чем полноценный парсер, но устойчивы и
+интерпретируемы.
+
+---
+
+## 10. Читаемость
+
+| Признак | Формула |
+|---|---|
+| `read_syllables_per_word` | Слоги / слова |
+| `read_polysyllable_rate` | Доля слов с 3+ слогами |
+| `read_flesch` | \(206{,}835 - 1{,}015 \cdot W/S - 84{,}6 \cdot Sy/W\) |
+| `read_fk_grade` | \(0{,}39 \cdot W/S + 11{,}8 \cdot Sy/W - 15{,}59\) |
+| `read_gunning_fog` | \(0{,}4 \cdot (W/S + 100 \cdot P/W)\) |
+| `read_smog` | \(1{,}043\sqrt{30P/S} + 3{,}129\) |
+| `read_coleman_liau` | \(0{,}0588L - 0{,}296S_{100} - 15{,}8\) |
+| `read_ari` | \(4{,}71 \cdot L/W + 0{,}5 \cdot W/S - 21{,}43\) |
+
+где \(W\) — слова, \(S\) — предложения, \(S_{100}\) — предложений на 100 слов,
+\(Sy\) — слоги, \(P\) — слова с 3+ слогами, \(L\) — буквы на 100 слов. Число
+слогов оценивается эвристически.
+
+**Зачем.** Классические метрики сложности. **Ограничение:** на коротких
+разговорных репликах формулы приблизительны; признаки помечены как
+предварительные.
+
+---
+
+## 11. Разговорные и игровые маркеры
+
+| Признак | Смысл |
+|---|---|
+| `dlg_placeholder_rate` | Доля плейсхолдеров `<PLAYER>` и т.п. |
+| `dlg_all_caps_rate` | Доля слов в верхнем регистре |
+| `dlg_ellipsis_rate` | Многоточия на предложение |
+| `dlg_repeated_punctuation_rate` | `!!`, `??`, `?!` на предложение |
+| `dlg_interjection_rate` | `oh`, `hmm`, `sigh`, `giggle` и т.п. |
+| `dlg_laughter_rate` | `haha`, `hehe`, `lol` |
+| `dlg_discourse_marker_rate` | `well`, `just`, `like`, `so` |
+| `dlg_hedge_rate` | `maybe`, `kind of`, `I guess` |
+| `dlg_intensifier_rate` | `very`, `really`, `so`, `too` |
+| `dlg_politeness_rate` | `please`, `thanks`, `sorry` |
+| `dlg_negation_rate` | Отрицания |
+| `dlg_contraction_rate` | Сокращения и апострофы |
+| `dlg_question_tag_rate` | `right?`, `isn't it?` |
+| `dlg_filler_rate` | `um`, `uh`, `hmm` |
+| `dlg_emotion_word_rate` | Небольшой список эмоциональных слов |
+| `dlg_first/second_person_rate` | Ориентация на себя/собеседника |
+| `dlg_one_word_sentence_rate` | Доля однословных предложений |
+
+**Зачем.** Это признаки, специфичные для игровых реплик: они отражают
+эмоциональность, разговорность и авторские привычки оформления. Многие из
+них недоступны в формальных корпусах, что делает корпус Stardew Valley
+интересным для стилометрии.
+
+---
+
+## 12. Символьные и POS n-граммы
+
+- **Символьные n-граммы:** TF-IDF по `char_wb`, 2–5 символов, `min_df=3`,
+  до 5000 признаков, L2-норма.
+- **POS n-граммы:** TF-IDF по последовательностям Universal POS, 1–3 тега,
+  `min_df=3`, до 500 признаков, L2-норма.
+
+**Зачем.** Символьные n-граммы ловят морфологию, пунктуацию и орфографию;
+POS n-граммы — синтаксические шаблоны. Оба блока устойчивее к теме, чем
+словарные n-граммы.
+
+---
+
+## 13. LSA
+
+LSA — не ручной признак, а способ понижения размерности TF-IDF:
+
+```text
+TF-IDF -> TruncatedSVD(k) -> L2-нормализация
+```
+
+Варианты: `word`, `word2`, `content`, `fw`, `char`, `pos`. Подробности — в
+`EXPERIMENTS.md` и `stylometry/lsa.py`.
+
+**Зачем.** LSA сжимает разреженное пространство и может выделить латентные
+оси вариабельности. Важно сравнивать варианты: LSA по всем словам может
+отражать тему, а LSA по POS и функциональным словам — стиль.
+
+---
+
+## 14. Что не используется и почему
+
+- **Лемматизация по умолчанию.** Инфлексия (время, число, степень) — часть
+  стиля; лемматизация её удаляет. Можно добавить как отдельный вариант LSA.
+- **Удаление стоп-слов в основных признаках.** Функциональные слова — ядро
+  стилометрии; удаление допустимо только для content-LSA.
+- **Удаление пунктуации.** Пунктуация — сильный стилометрический сигнал.
+- **Удаление коротких реплик.** Они различают персонажей (Jas, Vincent,
+  Krobus) и уже прошли ручную фильтрацию.
+- **SMOTE и синтез текста.** Может создать неестественные комбинации слов и
+  исказить стилометрический сигнал.
diff --git a/lab1/METRICS_AND_CLASSIFIERS.md b/lab1/METRICS_AND_CLASSIFIERS.md
new file mode 100644
index 0000000..218a1aa
--- /dev/null
+++ b/lab1/METRICS_AND_CLASSIFIERS.md
@@ -0,0 +1,894 @@
+# Метрики и классификаторы: методическое описание
+
+Документ предназначен для отчёта и презентации. Он объясняет, что показывает
+каждая метрика, как она вычисляется, как интерпретировать значения и чем
+отличаются классификаторы. Аудитория может не знать деталей машинного
+обучения, но предполагается, что базовые понятия выборки, признака и класса
+знакомы.
+
+Обозначения:
+
+- \(n\) — число объектов (реплик) в выборке;
+- \(C\) — число классов (в основном эксперименте \(C=34\));
+- \(y_i\) — истинный класс объекта \(i\);
+- \(\hat y_i\) — предсказанный класс;
+- \(c\) — индекс класса;
+- \(n_c\) — число объектов класса \(c\);
+- \(\hat p_{i,c}\) — вероятность, которую модель приписывает объекту \(i\)
+  и классу \(c\).
+
+---
+
+## 1. Постановка задачи
+
+Решается задача **многоклассовой классификации с одним правильным ответом**:
+каждая реплика относится ровно к одному персонажу. Модель получает вектор
+стилометрических признаков и должна выбрать один из 34 классов.
+
+Особенности, которые влияют на выбор метрик:
+
+1. **Дисбаланс классов.** Число реплик изменяется от 15 до 139. Модель, которая
+   всегда предсказывает самый частый класс, может иметь заметную долю
+   правильных ответов, но бесполезна для редких персонажей.
+2. **Много классов.** Случайное угадывание даёт около \(1/34 \approx 2{,}9\%\).
+   Поэтому полезны top-3 и top-5.
+3. **Короткие тексты.** Один объект — реплика длиной около 19 слов. Оценки
+   качества будут ниже, чем на длинных документах; это ожидаемо.
+4. **Один автор.** Задача измеряет различимость речевых масок, а не авторскую
+   атрибуцию.
+
+---
+
+## 2. Матрица ошибок
+
+**Матрица ошибок** (confusion matrix) — таблица \(C \times C\), где строка
+соответствует истинному классу, столбец — предсказанному. Элемент \(m_{ab}\) —
+число объектов класса \(a\), которые модель отнесла к классу \(b\).
+
+Для одного класса \(c\) удобно рассматривать задачу «класс \(c\) против
+остальных»:
+
+| | Предсказан \(c\) | Предсказан не \(c\) |
+|---|---:|---:|
+| Истинно \(c\) | \(TP_c\) (истинно положительные) | \(FN_c\) (ложно отрицательные) |
+| Истинно не \(c\) | \(FP_c\) (ложно положительные) | \(TN_c\) (истинно отрицательные) |
+
+Диагональ матрицы — правильные ответы. Вне диагонали — ошибки. По матрице
+видно не только «сколько ошибок», но и **какие персонажи перепутываются**.
+Это важно для содержательной интерпретации: например, ошибки между детьми
+(Jas и Vincent) или между нелюдьми (Krobus, Dwarf, Wizard) могут отражать
+сходство их речевых масок.
+
+---
+
+## 3. Метрики для одного класса
+
+### 3.1. Точность (precision)
+
+\[
+P_c = \frac{TP_c}{TP_c + FP_c}.
+\]
+
+Показывает, какая доля объектов, которые модель назвала классом \(c\),
+действительно принадлежит этому классу. Высокая точность означает мало
+ложных срабатываний.
+
+Пример: если модель 10 раз предсказала Abigail, и 7 раз правильно, точность
+равна \(7/10 = 0{,}7\).
+
+### 3.2. Полнота (recall)
+
+\[
+R_c = \frac{TP_c}{TP_c + FN_c}.
+\]
+
+Показывает, какую долю объектов класса \(c\) модель нашла. Высокая полнота
+означает мало пропусков.
+
+Пример: если в выборке 20 реплик Abigail, и модель правильно нашла 12,
+полнота равна \(12/20 = 0{,}6\).
+
+### 3.3. F-мера (F1)
+
+\[
+F1_c = \frac{2 P_c R_c}{P_c + R_c}.
+\]
+
+Гармоническое среднее точности и полноты. Почему гармоническое, а не
+арифметическое: оно сильнее наказывает за низкое значение любой из двух
+метрик. Если точность 1,0, а полнота 0,1, арифметическое среднее 0,55, а
+F-мера — около 0,18. Это соответствует интуиции: модель, которая находит
+только десятую часть класса, нельзя считать хорошей.
+
+F-мера полезна, когда важно и не пропускать объекты класса, и не приписывать
+к нему чужие. В нашей задаче это естественная метрика для каждого персонажа.
+
+### 3.4. Поддержка (support)
+
+\(n_c\) — число истинных объектов класса \(c\) в выборке. Support нужен,
+чтобы понимать, насколько надёжна оценка: F-мера по классу с 15 объектами
+имеет гораздо больший разброс, чем по классу с 139 объектами.
+
+---
+
+## 4. Способы усреднения метрик
+
+### 4.1. Macro-усреднение
+
+\[
+\text{macro-F1} = \frac{1}{C}\sum_{c=1}^{C} F1_c.
+\]
+
+Каждый класс получает одинаковый вес независимо от числа объектов. Это
+**основная метрика** в нашей задаче, потому что исследовательский вопрос
+касается всех 34 персонажей, включая редких.
+
+Недостаток: редкий класс с 15 репликами влияет на итог так же сильно, как
+класс с 139 репликами; поэтому macro-F1 может сильно колебаться между
+блоками перекрёстной проверки.
+
+### 4.2. Взвешенное усреднение
+
+\[
+\text{weighted-F1} = \sum_{c=1}^{C} \frac{n_c}{n} F1_c.
+\]
+
+Каждый класс взвешен по числу объектов. Метрика ближе к качеству на частых
+классах; её полезно приводить как дополнительную, но не как основную.
+
+### 4.3. Микро-усреднение
+
+Сначала суммируются \(TP\), \(FP\), \(FN\) по всем классам, затем считается
+общая F-мера. В задаче с одним правильным ответом micro-F1 **совпадает с
+долей правильных ответов** (accuracy). Поэтому отдельно его считать не нужно,
+но можно указать для полноты.
+
+### 4.4. Пример влияния усреднения
+
+Пусть модель хорошо различает Abigail, Alex и Haley, но почти не различает
+редких персонажей. Тогда:
+
+- доля правильных ответов и взвешенная F-мера будут относительно высокими;
+- macro-F1 будет заметно ниже;
+- F-мера по классам покажет, какие именно классы провалились.
+
+Именно поэтому в отчёте нужны macro-F1 **и** отчёт по классам.
+
+---
+
+## 5. Основные метрики
+
+### 5.1. Доля правильных ответов (accuracy)
+
+\[
+\text{Accuracy} = \frac{1}{n}\sum_{i=1}^{n} \mathbf{1}[y_i = \hat y_i].
+\]
+
+Доля объектов, для которых предсказание совпало с истинным классом.
+
+Плюсы: проста и интуитивна. Минусы: при дисбалансе может вводить в
+заблуждение. Пример из наших прогонов: модель `char + multinomial_nb` имеет
+долю правильных ответов около 9,7%, но сбалансированная доля правильных
+ответов 4,7% и macro-F1 0,028, потому что она часто предсказывает частые
+классы и почти не находит редкие.
+
+Случайный уровень: \(1/34 \approx 2{,}9\%\). Модель, всегда предсказывающая
+самый частый класс, даёт около \(139/2005 \approx 6{,}9\%\) правильных
+ответов, но
+macro-F1 около 0,004.
+
+### 5.2. Сбалансированная доля правильных ответов (balanced accuracy)
+
+\[
+\text{BalancedAccuracy} = \frac{1}{C}\sum_{c=1}^{C} R_c.
+\]
+
+Средняя полнота по классам. Каждый класс получает одинаковый вес, поэтому
+метрика устойчива к дисбалансу. Случайный уровень \(1/C\).
+
+Плюс: хорошо показывает, находит ли модель редкие классы. Минус: не учитывает
+точность (ложные срабатывания), поэтому её нужно дополнять macro-F1 и MCC.
+
+### 5.3. Macro-F1
+
+Основная метрика. Формула приведена выше. Интерпретация:
+
+- 0 — модель не находит ни одного класса;
+- \(1/C \approx 0{,}029\) — уровень случайного угадывания;
+- 0,10–0,20 — слабый, но заметный сигнал;
+- 0,30–0,50 — умеренное качество;
+- выше 0,50 — сильный результат для 34 классов и коротких реплик.
+
+Пороговые значения условны: они зависят от числа классов, длины текста и
+сложности задачи. В отчёте лучше сравнивать с базовой моделью и доверительными
+интервалами, а не с абстрактной шкалой.
+
+### 5.4. Коэффициент корреляции Мэтьюса (MCC)
+
+Для двух классов:
+
+\[
+\text{MCC} = \frac{TP \cdot TN - FP \cdot FN}
+{\sqrt{(TP+FP)(TP+FN)(TN+FP)(TN+FN)}}.
+\]
+
+Для многих классов используется обобщение через ковариацию матриц истинных и
+предсказанных индикаторов (так реализовано в scikit-learn). MCC принимает
+значения от \(-1\) до \(1\):
+
+- \(1\) — идеальные предсказания;
+- \(0\) — уровень случайного угадывания;
+- \(-1\) — полная инверсия классов.
+
+MCC удобен тем, что учитывает все четыре клетки матрицы ошибок и устойчив к
+дисбалансу. В отличие от F-меры, он одинаково штрафует за ложные
+срабатывания и пропуски.
+
+### 5.5. Каппа Коэна (Cohen's kappa)
+
+\[
+\kappa = \frac{p_o - p_e}{1 - p_e},
+\]
+
+где \(p_o\) — наблюдаемая доля согласия (accuracy), \(p_e\) — ожидаемая доля
+согласия при случайном предсказании с учётом распределения классов.
+
+Интерпретация:
+
+- \(\kappa = 0\) — согласие не лучше случайного;
+- \(\kappa = 1\) — полное согласие;
+- отрицательные значения — хуже случайного.
+
+Полезна как дополнение к доле правильных ответов: показывает, насколько модель превосходит
+случайное угадывание с тем же распределением классов. При сильном дисбалансе
+каппа может быть нестабильной; используйте вместе с macro-F1.
+
+### 5.6. Top-k
+
+\[
+\text{Top-}k = \frac{1}{n}\sum_{i=1}^{n}
+\mathbf{1}\left[y_i \in \text{Top-}k(\hat p_{i,\cdot})\right].
+\]
+
+Доля объектов, для которых истинный класс попал в \(k\) наиболее вероятных
+(или наиболее высоко оценённых) классов.
+
+При 34 классах top-1 может быть низким, но top-3 и top-5 показывают, что
+модель «чувствует» правильную область. Случайный уровень:
+
+- top-3: \(3/34 \approx 8{,}8\%\);
+- top-5: \(5/34 \approx 14{,}7\%\).
+
+Top-k особенно полезен для вероятностных моделей и моделей с решающей
+функцией. Для моделей без вероятностей ранжирование строится по значению
+решающей функции; это не калиброванные вероятности, но порядок классов
+корректен.
+
+### 5.7. Логарифмическая потеря (log loss)
+
+\[
+\text{LogLoss} = -\frac{1}{n}\sum_{i=1}^{n} \log \hat p_{i,y_i}.
+\]
+
+Оценивает качество вероятностей, а не только правильность ответа. Если модель
+уверена в неправильном классе, штраф растёт логарифмически.
+
+Случайный уровень для 34 классов: \(\ln 34 \approx 3{,}53\). Значения ниже
+лучше. В предварительных прогонах значения различаются: около 3,1 для
+объединённого набора с n-граммами и около 7,3 для одних ручных признаков.
+Чем ближе значение к 3,53, тем лучше калибровка; значения существенно выше
+означают, что модель часто уверена, но ошибается. Это не обязательно плохо
+для классификации, но важно для вероятностных выводов.
+
+Log loss определён только для моделей, выдающих вероятности. Для `linear_svc`
+он не считается.
+
+### 5.8. Оценка Брайера (Brier score)
+
+Для многоклассовой задачи:
+
+\[
+\text{Brier} = \frac{1}{n}\sum_{i=1}^{n}\sum_{c=1}^{C}
+\left(\hat p_{i,c} - \mathbf{1}[y_i = c]\right)^2.
+\]
+
+Средний квадрат отклонения предсказанных вероятностей от вектора с единицей
+в позиции истинного класса. Чем меньше, тем лучше. В отличие от log loss, Brier менее
+чувствителен к редким уверенным ошибкам, но тоже оценивает калибровку.
+
+Ориентир: для равномерного случайного распределения по 34 классам
+\(\text{Brier} \approx (C-1)/C \approx 0{,}971\). Значения выше этого
+означают, что модель хуже равномерного случайного распределения по качеству
+вероятностей.
+
+### 5.9. Матрица ошибок и отчёт по классам
+
+Матрица ошибок — не одна метрика, а инструмент анализа. В отчёте полезны:
+
+- **абсолютная матрица** — сколько реплик перепутано;
+- **нормированная по строкам матрица** — какая доля истинного класса ушла в
+  каждый предсказанный;
+- **самые частые пары ошибок** — какие классы чаще всего перепутываются.
+
+Отчёт по классам содержит точность, полноту, F-меру и support для каждого
+персонажа. Именно он показывает, что средние метрики могут скрывать провал
+редких классов.
+
+---
+
+## 6. Метрики верификации «свой/чужой»
+
+В верификации задача формулируется иначе: для тестовой реплики нужно решить,
+принадлежит ли она тому же персонажу, что и эталонный центроид. Это бинарная
+задача; положительный класс — «свой», отрицательный — «чужой».
+
+### 6.1. ROC-AUC
+
+Площадь под ROC-кривой. Интерпретация: вероятность того, что случайно взятый
+положительный объект получил более высокую оценку, чем случайно взятый
+отрицательный.
+
+- 0,5 — случайное ранжирование;
+- 1,0 — идеальное разделение;
+- ниже 0,5 — модель систематически ставит чужие объекты выше своих.
+
+ROC-AUC не зависит от выбранного порога, поэтому удобна для сравнения
+способов построения оценок.
+
+### 6.2. Равная ошибка (EER)
+
+Порог, при котором доля ложных принятий (FPR) равна доле ложных отклонений
+(FNR). Чем меньше EER, тем лучше. EER удобна как одна точка на ROC-кривой,
+но не заменяет всю кривую.
+
+### 6.3. TAR@FAR
+
+Доля истинных принятий при фиксированной доле ложных принятий. Например,
+TAR@FAR=1% показывает, какую долю «своих» можно принять, если разрешить не
+более 1% ложных принятий. В практических системах это часто важнее, чем EER.
+
+---
+
+## 7. Доверительные интервалы и статистическая значимость
+
+### 7.1. Зачем нужны интервалы
+
+Оценка качества по одной выборке — случайная величина. Если повторить
+разбиение, macro-F1 изменится. Поэтому в отчёте нужны не только точечные
+значения, но и интервалы.
+
+### 7.2. Бутстрэп (bootstrap)
+
+Алгоритм:
+
+1. Из \(n\) объектов случайно с возвращением выбираются \(n\) объектов.
+2. На этой выборке считается метрика.
+3. Шаги 1–2 повторяются \(B\) раз (обычно 1000).
+4. Берутся 2,5-й и 97,5-й процентили — это 95% доверительный интервал.
+
+Плюс: не требует предположений о распределении метрики. Минус: объекты
+считаются независимыми; если реплики внутри сцен коррелируют, интервал может
+быть слишком узким.
+
+### 7.3. Критерий Мак-Немара
+
+Сравнивает две модели на одних и тех же объектах. Пусть:
+
+- \(b\) — число объектов, где модель A правильна, а B ошибается;
+- \(c\) — число объектов, где A ошибается, а B правильна.
+
+При нулевой гипотезе «модели одинаковы» \(b\) и \(c\) — результаты
+\(b+c\) испытаний с вероятностью \(1/2\). Точный двусторонний критерий
+использует биномиальное распределение. Если \(b+c\) мало, критерий
+консервативен.
+
+### 7.4. Критерий Уилкоксона
+
+Парный непараметрический критерий для двух наборов оценок (например, macro-F1
+по блокам перекрёстной проверки). Проверяет, что медиана разностей не равна
+нулю. Не требует нормальности. При малом числе блоков мощность низкая.
+
+### 7.5. Критерий Фридмана
+
+Сравнивает три и более модели на одних и тех же блоках. Нулевая гипотеза —
+все модели одинаковы. При отклонении нулевой гипотезы можно использовать
+post-hoc критерий Неменьи и поправку Холма. В отчёте достаточно указать
+статистику, p-значение и поправку.
+
+### 7.6. Поправка Холма—Бонферрони
+
+При \(m\) одновременных проверках вероятность хотя бы одного ложного
+отклонения растёт. Поправка Холма упорядочивает p-значения и умножает их на
+\((m - \text{ранг})\). Она менее консервативна, чем классическая поправка
+Бонферрони, и широко используется.
+
+### 7.7. Размер эффекта
+
+p-значение отвечает на вопрос «есть ли различие», но не «насколько оно
+велико». Поэтому приводите разницу macro-F1 и её доверительный интервал.
+Разница в 1–2 процентных пункта при 34 классах и 2005 объектах может быть
+статистически значимой, но практически незначимой.
+
+---
+
+## 8. Метрики для LSA и кластеризации
+
+### 8.1. Объяснённая дисперсия
+
+`TruncatedSVD` раскладывает матрицу TF-IDF по компонентам. Доля объяснённой
+дисперсии компоненты \(j\):
+
+\[
+\text{EVR}_j = \frac{\sigma_j^2}{\sum_k \sigma_k^2},
+\]
+
+где \(\sigma_j\) — сингулярное число. Накопленная доля — сумма первых \(k\).
+Она показывает, какая доля вариабельности матрицы сохраняется при усечении.
+
+Важно: высокая объяснённая дисперсия не гарантирует полезность для
+классификации. Выбор \(k\) делается по внешней метрике (macro-F1) и
+интерпретируемости, а не только по графику объяснённой дисперсии.
+
+### 8.2. Ошибка восстановления
+
+Относительная ошибка \(\|X - X_k\|_F / \|X\|_F\), где \(X_k\) —
+приближение матрицы после SVD. Показывает, насколько потеря информации велика.
+Полезна как дополнительная диагностика.
+
+### 8.3. ARI, NMI, силуэт
+
+- **ARI** (Adjusted Rand Index) — согласие кластеризации с истинными классами
+  с поправкой на случай. 1 — идеальное совпадение, 0 — случайный уровень.
+- **NMI** (Normalized Mutual Information) — взаимная информация между
+  кластеризацией и классами, нормированная в [0,1].
+- **Силуэт** — насколько объект ближе к своему кластеру, чем к соседнему.
+  Значения от -1 до 1; выше лучше.
+- **Calinski–Harabasz** — отношение межкластерной дисперсии к внутрикластерной;
+  выше лучше.
+- **Davies–Bouldin** — среднее отношение внутрикластерного разброса к
+  межкластерному; ниже лучше.
+
+Кластеризация центроидов персонажей — предварительный анализ. ARI может быть
+низким даже при полезной структуре; не делайте из кластеризации сильных
+выводов.
+
+---
+
+## 9. Классификаторы
+
+### 9.1. Зачем несколько классификаторов
+
+Разные модели по-разному извлекают сигнал:
+
+- вероятностные модели (наивный Байес) хорошо работают с разреженными
+  частотными признаками;
+- линейные модели (логистическая регрессия, SVM) устойчивы при большом числе
+  признаков и малом числе объектов;
+- деревья и ансамбли ловят нелинейные взаимодействия;
+- метод ближайших соседей использует локальную структуру.
+
+Сравнение нескольких моделей нужно не для «гонки за процентом», а чтобы
+показать, что вывод не зависит от одного алгоритма. Если разные модели дают
+близкие результаты, вывод надёжнее.
+
+### 9.2. DummyClassifier (базовая модель)
+
+Предсказывает самый частый класс или случайный класс с учётом распределения.
+Не использует признаки. Нужен как нижняя граница: если модель не лучше
+Dummy, она не извлекла сигнал.
+
+В нашей задаче Dummy с самым частым классом даёт долю правильных ответов
+около 6,9%, но macro-F1 около 0,004. Это хороший пример того, почему долю
+правильных ответов нельзя использовать отдельно.
+
+### 9.3. Наивный байесовский классификатор (Multinomial Naive Bayes)
+
+Идея: применить теорему Байеса и предположить, что признаки условно
+независимы при данном классе:
+
+\[
+P(c \mid x) \propto P(c)\prod_{j=1}^{d} P(x_j \mid c).
+\]
+
+Для текстовых частот используется мультиномиальное распределение. Чтобы не
+было нулевых вероятностей, применяется сглаживание Лапласа:
+
+\[
+P(w \mid c) = \frac{N_{w,c} + \alpha}{N_c + \alpha V},
+\]
+
+где \(N_{w,c}\) — число вхождений слова \(w\) в класс \(c\), \(N_c\) — общее
+число слов класса, \(V\) — размер словаря, \(\alpha\) — параметр сглаживания.
+
+Плюсы:
+
+- очень быстрый;
+- хорошо работает на разреженных TF-IDF и частотных признаках;
+- устойчив при малом числе объектов;
+- легко объясняется.
+
+Минусы:
+
+- предположение о независимости признаков почти всегда ложно;
+- чувствителен к дисбалансу, если не использовать веса классов;
+- требует неотрицательных признаков; поэтому LSA-компоненты (они могут быть
+  отрицательными) в нашей реализации исключены;
+- вероятности плохо калиброваны.
+
+В нашем конвейере для MultinomialNB используются ручные неотрицательные
+признаки после MinMax-масштабирования, символьные и POS n-граммы. Это
+честная базовая модель, но не основной кандидат.
+
+### 9.4. Логистическая регрессия
+
+Линейная модель. Для каждого класса вычисляется линейная оценка:
+
+\[
+s_c(x) = w_c^\top x + b_c.
+\]
+
+Вероятности получаются через softmax:
+
+\[
+P(c \mid x) = \frac{\exp(s_c(x))}{\sum_{k=1}^{C}\exp(s_k(x))}.
+\]
+
+Обучение минимизирует перекрёстную энтропию:
+
+\[
+\mathcal{L} = -\frac{1}{n}\sum_{i=1}^{n}\log P(y_i \mid x_i)
++ \frac{1}{2C}\|w\|_2^2.
+\]
+
+Параметр \(C\) — обратная сила регуляризации: меньше \(C\) — сильнее
+регуляризация, проще модель. В нашей реализации используется L2 и
+`class_weight="balanced"`, чтобы редкие классы не игнорировались.
+
+Плюсы:
+
+- сильная и стабильная базовая модель для текста;
+- хорошо работает при \(d \gg n\);
+- коэффициенты интерпретируемы: видно, какие признаки увеличивают шанс
+  класса;
+- выдаёт вероятности.
+
+Минусы:
+
+- только линейные границы;
+- чувствительна к масштабу признаков (поэтому стандартизация);
+- при сильной мультиколлинеарности коэффициенты нестабильны;
+- вероятности могут быть плохо калиброваны.
+
+В наших прогонах логистическая регрессия — основная модель для сравнения
+наборов признаков.
+
+### 9.5. Линейный метод опорных векторов (Linear SVM)
+
+Ищет линейную границу, максимизирующую зазор между классами. Для
+неразделимых данных используется мягкий зазор:
+
+\[
+\min_{w,b,\xi} \frac{1}{2}\|w\|^2 + C\sum_{i=1}^{n}\xi_i
+\]
+
+при ограничениях \(y_i(w^\top x_i + b) \ge 1 - \xi_i\), \(\xi_i \ge 0\).
+Параметр \(C\) задаёт штраф за нарушения зазора.
+
+Плюсы:
+
+- часто очень силён на текстовых признаках;
+- устойчив к высокой размерности;
+- хорошо работает с разреженными матрицами;
+- класс-веса поддерживаются.
+
+Минусы:
+
+- не выдаёт вероятности; есть только решающая функция;
+- требует калибровки, если нужны вероятности;
+- чувствителен к выбору \(C\);
+- для top-k используем значения решающей функции, а не вероятности.
+
+В отчёте не сравнивайте log loss у SVM: он не определён без калибровки.
+
+### 9.6. Случайный лес (Random Forest)
+
+Ансамбль решающих деревьев. Каждое дерево обучается на бутстрэп-выборке, а
+при разбиении рассматривается случайное подмножество признаков. Итоговый
+класс — голосование деревьев (или среднее вероятностей).
+
+Плюсы:
+
+- ловит нелинейные взаимодействия;
+- устойчив к выбросам и масштабу признаков;
+- даёт permutation importance;
+- поддерживает веса классов.
+
+Минусы:
+
+- склонен к переобучению при большом числе признаков и малом числе объектов;
+- медленнее линейных моделей;
+- importance смещены в пользу коррелированных признаков;
+- вероятности могут быть плохо калиброваны.
+
+В нашей задаче лес — дополнительная проверка, а не основной кандидат:
+2005 объектов и 34 класса — небольшая выборка для леса.
+
+### 9.7. Сверхслучайные деревья (ExtraTrees)
+
+Похожи на случайный лес, но пороги разбиения выбираются случайно, а не
+оптимизируются. Это уменьшает дисперсию и ускоряет обучение, но может
+увеличить смещение. Полезны как альтернатива случайному лесу.
+
+### 9.8. Метод k ближайших соседей (kNN)
+
+Для объекта находится \(k\) ближайших объектов обучающей выборки; класс
+определяется голосованием (или взвешенным голосованием). В нашей реализации
+используется косинусное расстояние и веса, обратные расстоянию.
+
+Плюсы:
+
+- простая и интерпретируемая идея;
+- не делает предположений о форме границы;
+- хорошо работает, если признаки хорошо нормированы.
+
+Минусы:
+
+- проклятие размерности: в высоком измерении расстояния становятся похожими;
+- чувствителен к масштабу и шуму;
+- медленный на больших выборках;
+- не выдаёт содержательных коэффициентов.
+
+kNN полезен как локальная базовая модель, но в стилометрии обычно уступает
+линейным моделям.
+
+### 9.9. Градиентный бустинг (опционально)
+
+Последовательно строит деревья, каждое из которых исправляет ошибки
+предыдущих. Обычно силён на табличных данных, но на 2005 объектах и 34
+классах легко переобучается. В основной конфигурации не используется; может
+быть добавлен как дополнительная модель при аккуратной настройке.
+
+### 9.10. Дельта Барроуза
+
+Классический метод стилометрии, не являющийся классификатором в смысле
+машинного обучения.
+
+Алгоритм:
+
+1. Выбрать \(m\) наиболее частотных слов обучающей выборки (MFW).
+2. Для каждого документа вычислить относительные частоты \(f_{ij}\).
+3. z-нормировать каждый признак по обучающей выборке:
+   \[
+   z_{ij} = \frac{f_{ij} - \mu_j}{\sigma_j}.
+   \]
+4. Вычислить центроид каждого класса.
+5. Отнести документ к классу с минимальным расстоянием:
+   \[
+   \Delta_i = \frac{1}{m}\sum_{j=1}^{m}|z_{ij} - z_{cj}|
+   \]
+   (манхэттенская дельта) или косинусным расстоянием (Cosine Delta).
+
+Плюсы:
+
+- стандартная и хорошо объяснимая базовая модель;
+- не требует сложной настройки;
+- устойчив к теме, если выбраны частотные слова;
+- легко воспроизводится.
+
+Минусы:
+
+- на коротких репликах оценки частот шумные;
+- число MFW \(m\) нужно подбирать;
+- z-нормирование чувствительно к выбросам;
+- уступает ML-моделям при наличии большого числа признаков.
+
+В отчёте дельта нужна как независимая проверка: если ML-модели и дельта
+показывают сходную картину, вывод надёжнее.
+
+### 9.11. Сравнение классификаторов
+
+| Модель | Тип | Вероятности | Работа с разреженными признаками | Веса классов | Интерпретируемость |
+|---|---|---|---|---|---|
+| Dummy | базовая модель | да | не использует | нет | тривиальная |
+| MultinomialNB | вероятностная | да | хорошо | нет | средняя |
+| LogisticRegression | линейная | да | хорошо | да | высокая |
+| LinearSVC | линейная | нет | хорошо | да | средняя |
+| RandomForest | ансамбль деревьев | да | удовлетворительно | да | средняя |
+| ExtraTrees | ансамбль деревьев | да | удовлетворительно | да | средняя |
+| kNN | локальная | да | удовлетворительно | нет | низкая |
+| Delta | метрическая | нет/условно | хорошо | нет | высокая |
+
+---
+
+## 10. Схема валидации
+
+### 10.1. Обучение и тест
+
+Модель настраивается на обучающей выборке и проверяется на тестовой. Нельзя
+оценивать модель на тех же объектах, на которых она обучалась: она может
+запомнить их и показать завышенное качество (переобучение).
+
+### 10.2. Перекрёстная проверка
+
+При перекрёстной проверке (cross-validation) выборка делится на \(k\)
+блоков. Модель обучается на \(k-1\) блоках и проверяется на оставшемся;
+процедура повторяется \(k\) раз. Итог — среднее и разброс метрик.
+
+### 10.3. Стратификация
+
+При стратификации сохраняется доля классов в каждом блоке. Это важно при
+дисбалансе: иначе редкий класс может вообще не попасть в тестовый блок.
+В нашей задаче 5 блоков и минимум 15 реплик на класс дают около 3 тестовых
+реплик на класс.
+
+### 10.4. Повторность
+
+Одного разбиения недостаточно: результат зависит от случайного деления.
+Повторная перекрёстная проверка (repeated cross-validation) выполняет
+несколько разбиений с разными seed и даёт более устойчивую оценку.
+
+### 10.5. Вложенная перекрёстная проверка
+
+Гиперпараметры нельзя настраивать на тестовых блоках. Вложенная схема:
+
+- внешние блоки — оценка качества;
+- внутренние блоки — настройка гиперпараметров.
+
+Такой протокол дороже, но честнее.
+
+### 10.6. Утечка данных
+
+Утечка (data leakage) — ситуация, когда информация из тестовой выборки
+попадает в обучение. Типичные причины:
+
+- TF-IDF или SVD обучены на всём корпусе до кросс-валидации;
+- отбор признаков выполнен по всей выборке;
+- масштабирование выполнено до разделения;
+- почти одинаковые тексты есть и в обучении, и в тесте.
+
+В нашем конвейере все обучаемые преобразования находятся внутри `Pipeline` и
+настраиваются только на обучающих блоках. Ручные признаки и POS-теги не
+используют метки, поэтому их можно считать заранее.
+
+### 10.7. Дисбаланс
+
+Основные способы:
+
+- веса классов (`class_weight="balanced"`);
+- macro-F1 и сбалансированную долю правильных ответов;
+- отчёт по классам;
+- стратификация.
+
+Не используйте SMOTE и другие методы синтеза текста без крайней
+необходимости: они могут создать искусственные комбинации слов и исказить
+стилометрический сигнал.
+
+---
+
+## 11. Как объяснять результаты на защите
+
+1. **Сначала постановка.** 2005 реплик, 34 персонажа, один автор; задача —
+   различить речевые маски, а не авторов.
+2. **Затем данные и ограничения.** Короткие реплики, дисбаланс, ручная
+   фильтрация, отсутствие сцен.
+3. **Затем признаки.** Функциональные слова, POS, символьные n-граммы, LSA;
+   почему именно они.
+4. **Затем валидация.** 5-блочная стратифицированная перекрёстная проверка,
+   повторность, вложенная настройка, контроль утечек.
+5. **Затем метрики.** Macro-F1 как основная; доля правильных ответов как дополнительная;
+   top-3/top-5; F-мера по классам; матрица ошибок.
+6. **Затем результаты.** Сравнение с базовой моделью, доверительные интервалы,
+   статистические тесты.
+7. **Затем интерпретация.** Какие персонажи различаются, какие
+   перепутываются, что это говорит о речевых масках.
+8. **Затем ограничения и улучшения.** Больше данных, агрегация реплик,
+   тематический контроль, более точный частеречный теггер, сцены.
+
+### 11.1. Пример корректной формулировки
+
+> «На 2005 репликах и 34 классах модель логистической регрессии по
+> объединённому набору стилометрических признаков, символьных и частеречных
+> n-грамм достигает macro-F1 0,18 при 5-блочной стратифицированной
+> перекрёстной проверке. Случайный уровень для 34 классов
+> составляет 0,029; top-3 равен 0,34 при случайном уровне 0,088. Это
+> указывает на статистически различимый, но не абсолютный сигнал. Основной
+> вклад дают символьные и частеречные n-граммы; LSA по всем словам может
+> частично отражать тему, поэтому мы сравниваем её с LSA по POS и
+> функциональным словам».
+
+### 11.2. Чего избегать
+
+- «Доля правильных ответов 10%, значит модель плохая» — без сравнения с
+  базовой моделью и macro-F1 это не вывод.
+- «Модель различает авторов» — автор один; различаются персонажи.
+- «LSA выявил стиль» — LSA может выявить тему; нужен контроль.
+- «p < 0,05, значит различие важное» — нужен размер эффекта и интервал.
+- «Кластеризация доказала группы» — кластеризация носит предварительный характер.
+
+---
+
+## 12. Краткая версия для слайдов
+
+### 12.1. Метрики
+
+| Метрика | Формулировка для слайда | Ориентир |
+|---|---|---|
+| Macro-F1 | Средняя F-мера по 34 классам; все персонажи важны одинаково | Случайно около 0,029 |
+| Сбалансированная доля правильных ответов | Средняя полнота по классам; показывает, находит ли модель редких персонажей | Случайно около 0,029 |
+| Доля правильных ответов | Доля совпавших предсказаний; при дисбалансе может вводить в заблуждение | Случайно около 0,029 |
+| MCC | Корреляция истинных и предсказанных меток от -1 до 1; устойчив к дисбалансу | 0 — случайно |
+| Каппа Коэна | Согласие с поправкой на случайное угадывание | 0 — случайно |
+| Top-3 | Истинный персонаж попал в тройку лучших ответов модели | Случайно около 0,088 |
+| Логарифмическая потеря | Качество вероятностей; штраф за уверенную ошибку | Случайно около 3,53 |
+| Оценка Брайера | Квадратичная ошибка вероятностей | Случайно около 0,971 |
+| ROC-AUC | Для верификации «свой/чужой»: качество ранжирования без выбора порога | 0,5 — случайно |
+| EER | Порог, где доля ложных принятий равна доле ложных отклонений | Чем меньше, тем лучше |
+| Бутстрэп-интервал | Диапазон, в котором с вероятностью 95% лежит метрика | Уже — устойчивее |
+| Мак-Немар | Парный тест: различаются ли две модели на одних и тех же репликах | p < 0,05 — есть различие |
+
+### 12.2. Классификаторы
+
+| Модель | Формулировка для слайда | Сильная сторона | Ограничение |
+|---|---|---|---|
+| Dummy | Нижняя граница: всегда предсказывает частый класс | Простота | Не использует признаки |
+| MultinomialNB | Быстрый вероятностный baseline для частотных признаков | Скорость, устойчивость | Независимость признаков, дисбаланс |
+| LogisticRegression | Линейная модель с вероятностями и интерпретируемыми коэффициентами | Стабильность, интерпретируемость | Только линейные границы |
+| LinearSVC | Линейная модель с максимальным зазором | Сильна на текстовых признаках | Нет калиброванных вероятностей |
+| RandomForest | Ансамбль деревьев, ловит нелинейные взаимодействия | Нелинейность, importance | Переобучение на малой выборке |
+| kNN | Локальная модель по близости объектов | Простая идея | Проклятие размерности |
+| Delta | Классическое расстояние до центроида по частотным словам | Независимый стилометрический baseline | Шум на коротких репликах |
+
+### 12.3. Одна фраза для вывода
+
+> «Качество оценивалось по macro-F1 с 95% бутстрэп-интервалом; модели
+> сравнивались точным критерием Мак-Немара, а не только по доле правильных
+> ответов. Устойчивость вывода проверяется сравнением нескольких
+> классификаторов: если разные модели дают близкие значения macro-F1 и top-3,
+> результат не является артефактом одного алгоритма».
+
+---
+
+## 13. Глоссарий
+
+| Русский термин | Английский термин | Краткое значение |
+|---|---|---|
+| Признак | feature | Измеримая характеристика объекта |
+| Класс | class | Один из вариантов ответа (персонаж) |
+| Обучающая выборка | training set | Объекты, на которых модель учится |
+| Тестовая выборка | test set | Объекты, на которых измеряется качество |
+| Перекрёстная проверка | cross-validation | Многократное деление на обучение и тест |
+| Блок | fold | Одна часть при перекрёстной проверке |
+| Стратификация | stratification | Сохранение долей классов в блоках |
+| Гиперпараметр | hyperparameter | Параметр модели, задаваемый до обучения |
+| Переобучение | overfitting | Запоминание обучающих данных вместо обобщения |
+| Утечка данных | data leakage | Информация из теста попадает в обучение |
+| Матрица ошибок | confusion matrix | Таблица истинных и предсказанных классов |
+| Точность | precision | Доля правильных среди предсказанных классом |
+| Полнота | recall | Доля найденных среди истинных класса |
+| F-мера | F1 score | Гармоническое среднее точности и полноты |
+| Macro-F1 | macro F1 | Среднее F1 по классам с равным весом |
+| Доля правильных ответов | accuracy | Доля совпавших предсказаний |
+| Сбалансированная доля | balanced accuracy | Средняя полнота по классам |
+| MCC | Matthews correlation coefficient | Корреляция истинных и предсказанных меток |
+| Каппа Коэна | Cohen's kappa | Согласие с поправкой на случай |
+| Top-k | top-k accuracy | Попал ли истинный класс в k лучших |
+| Логарифмическая потеря | log loss | Качество вероятностей |
+| Оценка Брайера | Brier score | Квадратичная ошибка вероятностей |
+| ROC-AUC | ROC AUC | Площадь под ROC-кривой |
+| EER | equal error rate | Точка равенства FPR и FNR |
+| Бутстрэп | bootstrap | Выборка с возвращением для интервалов |
+| Мак-Немар | McNemar test | Парный тест двух моделей |
+| Уилкоксон | Wilcoxon test | Парный непараметрический тест |
+| Фридман | Friedman test | Сравнение нескольких моделей |
+| LSA | latent semantic analysis | TF-IDF + сингулярное разложение |
+| TF-IDF | term frequency — inverse document frequency | Взвешивание слов |
+| SVD | singular value decomposition | Сингулярное разложение |
+| ARI | adjusted Rand index | Согласие кластеризации с классами |
+| NMI | normalized mutual information | Взаимная информация кластеризации и классов |
+| Силуэт | silhouette | Качество кластерной структуры |
+| Наивный Байес | naive Bayes | Вероятностная модель с независимостью признаков |
+| Логистическая регрессия | logistic regression | Линейная вероятностная модель |
+| Метод опорных векторов | support vector machine | Линейная модель с максимизацией зазора |
+| Случайный лес | random forest | Ансамбль решающих деревьев |
+| kNN | k-nearest neighbours | Классификация по ближайшим объектам |
+| Дельта Барроуза | Burrows' Delta | Метрика расстояния в стилометрии |
diff --git a/lab1/PRESENTATION_OUTLINE.md b/lab1/PRESENTATION_OUTLINE.md
new file mode 100644
index 0000000..6e98740
--- /dev/null
+++ b/lab1/PRESENTATION_OUTLINE.md
@@ -0,0 +1,228 @@
+# Структура презентации
+
+Ниже — рекомендуемая последовательность слайдов. Она рассчитана на аудиторию,
+которая знает базовые понятия машинного обучения, но не знает деталей
+стилометрии. Формулировки можно сокращать, но логику лучше сохранить.
+
+---
+
+## Слайд 1. Тема и исследовательский вопрос
+
+**Заголовок:** Различимы ли речевые маски персонажей Stardew Valley?
+
+**Ключевые тезисы:**
+
+- 2005 реплик, 34 персонажа, один автор.
+- Задача — не авторская атрибуция, а классификация речевых масок.
+- Вопрос: можно ли различить персонажей по стилометрическим признакам
+  отдельных реплик?
+
+**Визуализация:** примеры реплик 3–4 контрастных персонажей (Abigail, Jas,
+Krobus, Wizard).
+
+---
+
+## Слайд 2. Корпус и его ограничения
+
+**Ключевые тезисы:**
+
+- 2005 реплик, 34 класса, около 38 тыс. словоупотреблений.
+- Средняя длина — 19 слов, медиана — 17.
+- Классы неравномерны: 15–139 реплик.
+- Корпус прошёл ручную фильтрацию коротких и общих фраз; правила не
+  восстановлены.
+- Точных дублей реплик нет.
+
+**Визуализация:** горизонтальная диаграмма распределения классов; гистограмма
+длин реплик.
+
+---
+
+## Слайд 3. Почему это сложная задача
+
+**Ключевые тезисы:**
+
+- Один автор → нельзя использовать традиционную авторскую атрибуцию.
+- Короткие тексты → признаки словарного богатства и читаемости нестабильны.
+- Дисбаланс → accuracy обманчива, нужна macro-F1.
+- 34 класса → случайный уровень 2,9%, top-3 — 8,8%.
+- LSA может выучить тему, а не стиль.
+
+**Визуализация:** таблица «случайный уровень vs ожидаемый сигнал».
+
+---
+
+## Слайд 4. Предобработка и NLTK
+
+**Ключевые тезисы:**
+
+- Нормализация Unicode и плейсхолдеров.
+- Токенизация, разбиение на предложения, частеречная разметка NLTK.
+- PTB → Universal POS.
+- Не удаляем стоп-слова, пунктуацию и короткие реплики: это стилометрический
+  сигнал.
+- Не лемматизируем по умолчанию: инфлексия — тоже стиль.
+
+**Визуализация:** схема «реплика → токены → POS → признаки».
+
+---
+
+## Слайд 5. Стилометрические признаки
+
+**Ключевые тезисы:**
+
+- 454 ручных признака в 11 группах.
+- Функциональные слова — ядро классической стилометрии.
+- POS и символьные n-граммы — устойчивы к теме.
+- Пунктуация и разговорные маркеры — важны для игровых реплик.
+- Словарное богатство и читаемость — предварительные признаки.
+
+**Визуализация:** таблица групп признаков с примерами; словарь признаков в
+приложении.
+
+---
+
+## Слайд 6. LSA
+
+**Ключевые тезисы:**
+
+- LSA = TF-IDF → усечённое сингулярное разложение → L2-нормализация.
+- Варианты: слова, содержание без стоп-слов, функциональные слова,
+  символьные n-граммы, POS.
+- Число компонент выбирается по объяснённой дисперсии и внешней метрике.
+- LSA по всем словам может отражать тему; сравнение с POS/функциональными
+  словами обязательно.
+- Обучение LSA — только внутри блоков перекрёстной проверки.
+
+**Визуализация:** график объяснённой дисперсии; 2D PCA/t-SNE; примеры
+top-терминов компонент.
+
+---
+
+## Слайд 7. Схема валидации
+
+**Ключевые тезисы:**
+
+- 5-блочная стратифицированная перекрёстная проверка.
+- Повторность: 3–5 повторов.
+- Вложенная настройка гиперпараметров: внешние блоки — оценка, внутренние —
+  настройка.
+- Все обучаемые преобразования — внутри блоков.
+- Контроль утечек: TF-IDF, SVD, масштабирование, отбор признаков.
+
+**Визуализация:** схема outer/inner CV.
+
+---
+
+## Слайд 8. Метрики
+
+**Ключевые тезисы:**
+
+- Основная метрика — macro-F1: все 34 класса равноправны.
+- Дополнительно: сбалансированная доля правильных ответов, MCC, каппа,
+  top-3/top-5.
+- Accuracy приводится, но не используется как единственная метрика.
+- 95% бутстрэп-интервалы.
+- Парные сравнения: Мак-Немар; несколько моделей — Фридман и Холм.
+
+**Визуализация:** таблица метрик с формулами в приложении; пример «почему
+accuracy обманывает» (Dummy: 6,9% accuracy, macro-F1 0,004).
+
+---
+
+## Слайд 9. Классификаторы
+
+**Ключевые тезисы:**
+
+- Dummy — нижняя граница.
+- MultinomialNB — быстрый вероятностный baseline.
+- LogisticRegression — основной кандидат: линейная модель, вероятности,
+  интерпретируемые коэффициенты.
+- LinearSVC — сильная линейная альтернатива без вероятностей.
+- RandomForest/ExtraTrees — нелинейная проверка.
+- kNN — локальный baseline.
+- Delta Барроуза — классическая стилометрическая проверка.
+
+**Визуализация:** таблица «модель — сильная сторона — ограничение».
+
+---
+
+## Слайд 10. Результаты
+
+**Ключевые тезисы:**
+
+- Предварительные прогоны: macro-F1 около 0,18 для объединённого набора,
+  top-3 около 0,34.
+- Случайный уровень: macro-F1 0,029, top-3 0,088.
+- Символьные и POS n-граммы дают основной прирост.
+- LSA сама по себе слабее; в объединении даёт небольшой прирост.
+- Редкие классы проваливаются; смотреть F-меру по классам.
+
+**Визуализация:** таблица результатов; матрица ошибок; диаграмма F-меры по классам.
+
+---
+
+## Слайд 11. Стиль или тема?
+
+**Ключевые тезисы:**
+
+- Сравнение `lsa_word`/`lsa_content` с `lsa_pos`/`lsa_fw`/`lsa_char`.
+- Если word/content значительно лучше — вклад темы велик.
+- Если POS/FW сопоставимы — стиль и синтаксис информативны.
+- Маскировка содержательных слов и тематический контроль — следующий шаг.
+
+**Визуализация:** сравнительная диаграмма macro-F1 по вариантам LSA.
+
+---
+
+## Слайд 12. Ошибки и интерпретация
+
+**Ключевые тезисы:**
+
+- Самые частые пары ошибок.
+- Персонажи с малым support.
+- Дети, нелюди, персонажи с похожей ролью.
+- Влияние длины реплик.
+- Анализ по полу, возрасту, виду.
+
+**Визуализация:** самые частые пары ошибок; F-мера по подгруппам.
+
+---
+
+## Слайд 13. Ограничения
+
+**Ключевые тезисы:**
+
+- Один автор.
+- Короткие тексты.
+- Ручная фильтрация корпуса.
+- Дисбаланс классов.
+- Нет идентификаторов сцен.
+- NLTK-теггер ошибается на разговорной речи.
+- LSA может отражать тему.
+- Кластеризация и t-SNE — предварительный анализ.
+
+---
+
+## Слайд 14. Выводы и дальнейшая работа
+
+**Ключевые тезисы:**
+
+- Речевые маски персонажей статистически различимы, но сигнал умеренный.
+- Основной вклад — символьные и частеречные n-граммы; LSA полезна как
+  дополнение, но требует контроля темы.
+- Для усиления: больше данных, агрегация реплик в псевдодокументы,
+  тематический контроль, более точный частеречный теггер, учёт сцен.
+- Воспроизводимость: код, конфигурации, seed, метаданные запусков.
+
+**Визуализация:** итоговая таблица «что сделано — что показано — что дальше».
+
+---
+
+## Приложение
+
+- Полный словарь признаков.
+- Формулы метрик.
+- Сетки гиперпараметров.
+- Примеры ошибок.
+- Команды запуска.
diff --git a/lab1/README.md b/lab1/README.md
new file mode 100644
index 0000000..981fc6b
--- /dev/null
+++ b/lab1/README.md
@@ -0,0 +1,231 @@
+# Стилометрия реплик Stardew Valley
+
+Проект для лабораторной работы по автоматическому анализу текстов. Задача —
+проверить, можно ли стилометрически различить персонажей игры, если все реплики
+написаны одним автором. Это не авторская атрибуция, а классификация
+**речевых масок персонажей** (character idiolect): автор намеренно наделяет
+героев разной лексикой, синтаксисом, междометиями и пунктуацией.
+
+Основная единица анализа — **одна реплика**. В текущем корпусе 2005 реплик,
+34 персонажа. Классы неравномерны: от 15 реплик у Mister Qi до 139 у Abigail.
+Поэтому основная метрика — macro-F1, а не доля правильных ответов.
+
+## Быстрый старт
+
+```bash
+cd lab1
+
+# 1. Установить зависимости (удобно в виртуальном окружении)
+python -m venv .venv
+source .venv/bin/activate
+pip install -r requirements.txt
+
+# 2. Скачать ресурсы NLTK в локальный каталог lab1/nltk_data
+python scripts/download_nltk_data.py
+
+# 3. Проверить корпус
+python -m stylometry.cli audit
+
+# 4. Построить ручные стилометрические признаки
+python -m stylometry.cli features
+
+# 5. Запустить кросс-валидацию
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --repeats 5
+```
+
+Если виртуальное окружение уже создано в корне проекта
+(`/home/mermutef/PycharmProjects/nlp/.venv`), команды можно запускать так:
+
+```bash
+cd lab1
+../.venv/bin/python -m stylometry.cli audit
+```
+
+Вместо `python -m stylometry.cli` можно использовать `python main.py`:
+
+```bash
+python main.py audit
+python main.py run --preset stylometry_pos_lsa --model logreg --repeats 5
+```
+
+## Что уже реализовано
+
+| Компонент | Модуль | Назначение |
+|---|---|---|
+| Чтение корпуса | `stylometry/io.py` | CSV в cp1251, XLSX, проверка колонок, SHA-256 |
+| Нормализация | `stylometry/cleaning.py` | Unicode, плейсхолдеры, пробелы |
+| Токенизация и POS | `stylometry/tokenization.py` | NLTK: `word_tokenize`, `sent_tokenize`, `pos_tag`; резервный режим |
+| Ручные признаки | `stylometry/features/` | длина, лексика, словарное богатство, символы, пунктуация, функциональные слова, местоимения, POS, синтаксис, читаемость, разговорные маркеры |
+| LSA | `stylometry/lsa.py` | TF-IDF + `TruncatedSVD` для word/content/fw/char/pos |
+| Конвейеры | `stylometry/pipelines.py` | объединение блоков признаков, классификаторы, сетки гиперпараметров |
+| Кросс-валидация | `stylometry/cv.py` | повторная стратифицированная и вложенная CV |
+| Метрики | `stylometry/evaluation.py` | macro-F1, сбалансированная доля правильных ответов, MCC, каппа, top-k, логарифмическая потеря, оценка Брайера, бутстрэп |
+| Статистика | `stylometry/stats.py` | Мак-Немар, Уилкоксон, Фридман, Холм |
+| Дельта Барроуза | `stylometry/delta.py` | MFW, z-нормирование, манхэттенское и косинусное расстояние |
+| Верификация | `stylometry/verification.py` | «тот же персонаж / другой персонаж», ROC-AUC, EER, TAR@FAR |
+| Кластеризация | `stylometry/clustering.py` | иерархическая и k-means, ARI, NMI, силуэт |
+| Визуализация | `stylometry/plotting.py` | распределение классов, scree, матрица ошибок, PCA, t-SNE |
+
+## Команды
+
+Все команды выполняются из каталога `lab1`.
+
+```bash
+# Аудит: отчёт data/processed/audit.md и таблица lines.csv
+python -m stylometry.cli audit
+
+# Матрица ручных признаков: features/stylometric.csv и словарь признаков
+python -m stylometry.cli features
+
+# LSA одного варианта
+python -m stylometry.cli lsa --variant word --components 100
+python -m stylometry.cli lsa --variant pos  --components 50
+python -m stylometry.cli lsa --variant char --components 50
+
+# Одна модель
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --repeats 5
+python -m stylometry.cli run --preset stylometry_pos     --model linear_svc --repeats 5
+python -m stylometry.cli run --preset stylometry         --model random_forest --repeats 3
+
+# Сравнение классификаторов
+python -m stylometry.cli benchmark --preset stylometry_pos_lsa \
+    --models logreg,linear_svc,random_forest,multinomial_nb,knn --repeats 3
+
+# Вложенная кросс-валидация с настройкой гиперпараметров
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --nested
+
+# Поочерёдное исключение групп ручных признаков
+python -m stylometry.cli ablate --preset stylometry_pos_lsa --model logreg --repeats 3
+
+# Перебор вариантов LSA и числа компонент
+python -m stylometry.cli lsa-sweep --variants word,pos,char --components 50,100,200
+
+# Дельта Барроуза
+python -m stylometry.cli delta --features 100,200,500 --metrics manhattan,cosine
+
+# Верификация «свой/чужой»
+python -m stylometry.cli verify --preset stylometry_pos_lsa
+
+# Кластеризация персонажей
+python -m stylometry.cli cluster --preset stylometry_pos_lsa
+
+# Метрики по группам длины реплик (по сохранённому запуску)
+python -m stylometry.cli length-report --run-dir results/<каталог_запуска>
+
+# Парное сравнение двух запусков: Мак-Немар и бутстрэп-разница macro-F1
+python -m stylometry.cli compare \
+    --run-a results/<каталог_запуска_A> \
+    --run-b results/<каталог_запуска_B>
+
+# Метрики по полу, возрасту и виду персонажей
+python -m stylometry.cli subgroup-report \
+    --run-dir results/<каталог_запуска> --group gender
+```
+
+Основные готовые наборы признаков (`--preset`):
+
+| Набор | Содержимое |
+|---|---|
+| `stylometry` | только ручные стилометрические признаки |
+| `stylometry_pos` | ручные признаки + символьные и POS n-граммы |
+| `stylometry_pos_lsa` | то же + LSA по словам и POS (основной набор) |
+| `all` | все блоки, включая все варианты LSA |
+| `char` | только символьные n-граммы |
+| `pos` | только POS n-граммы |
+| `fw` | функциональные слова и местоимения |
+| `lsa_word`, `lsa_pos` | только соответствующий вариант LSA |
+| `length` | только признаки длины (нижняя граница) |
+
+## Конфигурация
+
+Файл `configs/default.toml` задаёт параметры по умолчанию:
+
+- `[cv]` — число блоков, повторов и seed;
+- `[experiment]` — набор признаков и классификатор;
+- `[features]` — диапазоны n-грамм, число компонент LSA, `min_df`,
+  масштабирование.
+
+Параметры командной строки имеют приоритет над конфигурацией. Например:
+
+```bash
+python -m stylometry.cli run --config configs/default.toml \
+    --preset stylometry_pos_lsa --model logreg --repeats 3
+```
+
+## Основные документы
+
+- **`RUNBOOK.md`** — пошаговая инструкция для запуска полной серии
+  экспериментов: установка, проверка окружения, все команды, упаковка
+  результатов и типичные ошибки.
+- **`EXPERIMENTS.md`** — подробный протокол экспериментов: гипотезы, признаки,
+  схема валидации, матрица экспериментов, команды запуска, анализ ошибок,
+  ограничения и чек-лист для следующего исследователя.
+- **`FEATURES.md`** — состав стилометрических признаков, формулы и обоснование
+  выбора; дополняет машинно-читаемый `features/feature_dictionary.csv`.
+- **`METRICS_AND_CLASSIFIERS.md`** — методическое описание метрик и
+  классификаторов: что показывает каждая метрика, как она считается, как
+  интерпретировать значения, чем отличаются модели и как объяснять это на
+  защите.
+- **`PRESENTATION_OUTLINE.md`** — рекомендуемая структура презентации на
+  14 слайдов с ключевыми тезисами и предложениями по визуализации.
+
+## Структура проекта
+
+```text
+lab1/
+├── data/
+│   ├── dialogues.csv
+│   ├── dialogues.xlsx
+│   ├── stats_per_character.xlsx
+│   ├── metadata/characters.csv # пол, возраст, вид, роль
+│   └── processed/              # lines.csv, audit.md, tokenized.pkl
+├── features/                   # матрицы признаков и LSA
+├── stylometry/                 # основной пакет
+├── experiments/                # место для отдельных сценариев
+├── configs/default.toml
+├── results/                    # результаты запусков
+├── tests/                      # pytest
+├── scripts/download_nltk_data.py
+├── requirements.txt
+├── README.md
+├── EXPERIMENTS.md
+└── METRICS_AND_CLASSIFIERS.md
+```
+
+## Воспроизводимость
+
+- Корпус читается из `data/dialogues.csv` в кодировке cp1251. SHA-256 файла
+  сохраняется в отчёте и метаданных запуска.
+- Кэш токенизации `data/processed/tokenized.pkl` проверяется по SHA-256
+  исходного файла; при изменении корпуса пересобирается.
+- Все обучаемые преобразования (TF-IDF, SVD, масштабирование, отбор признаков)
+  настраиваются только на обучающем блоке кросс-валидации. Ручные признаки и
+  POS-теги не используют метки, поэтому могут считаться заранее.
+- Для каждого запуска сохраняются конфигурация, метрики по блокам, pooled
+  out-of-fold предсказания, матрица ошибок, отчёт по классам и графики.
+- Случайность фиксируется параметром `--seed` (по умолчанию 42).
+
+## Ограничения
+
+- Реплики короткие: в среднем около 19 слов. Признаки словарного богатства и
+  читаемости на таких текстах нестабильны; они помечены как предварительные.
+- Все тексты написаны одним автором; результат характеризует различимость
+  персонажей, а не авторский стиль.
+- Корпус прошёл ручную фильтрацию: короткие и общие фразы удалены. Точные
+  правила очистки не воспроизводимы, это ограничение зафиксировано в отчёте.
+- Классы неравномерны (15–139 реплик). Основная метрика — macro-F1; по редким
+  классам доверительные интервалы широкие.
+- NLTK-теггер обучен на внешнем корпусе и может ошибаться на разговорных
+  конструкциях и коротких фрагментах.
+- LSA по всем словам может отражать тему, а не стиль. Поэтому в протоколе
+  предусмотрено сравнение `lsa_word` с `lsa_pos`, `lsa_fw` и `lsa_char`.
+
+## Тесты
+
+```bash
+cd lab1
+python -m pytest -q
+```
+
+Тесты проверяют чтение корпуса, нормализацию, NLTK-разметку, признаки, LSA,
+дельту, метрики и кросс-валидацию.
diff --git a/lab1/RUNBOOK.md b/lab1/RUNBOOK.md
new file mode 100644
index 0000000..8a3a74c
--- /dev/null
+++ b/lab1/RUNBOOK.md
@@ -0,0 +1,686 @@
+# Инструкция по запуску экспериментов
+
+Документ предназначен для коллеги, который будет запускать серию
+экспериментов на своей машине. Все команды выполняются из каталога `lab1`.
+Результаты нужно упаковать и передать обратно одним архивом.
+
+Если вы впервые открыли проект, начните с разделов 1–4. Если окружение уже
+готово, переходите к разделу 5.
+
+---
+
+## 1. Что это за проект
+
+Исследуется, можно ли стилометрически различить 34 персонажа Stardew Valley
+по отдельным репликам, если все реплики написаны одним автором. Это
+многоклассовая классификация речевых масок персонажей, а не авторская
+атрибуция.
+
+Основные характеристики:
+
+- 2005 реплик;
+- 34 класса;
+- около 38 тыс. словоупотреблений;
+- средняя длина реплики около 19 слов;
+- классы неравномерны: от 15 до 139 реплик;
+- основная метрика — macro-F1;
+- схема валидации — повторная стратифицированная перекрёстная проверка.
+
+Полный протокол: `EXPERIMENTS.md`.
+Описание признаков: `FEATURES.md`.
+Описание метрик и классификаторов: `METRICS_AND_CLASSIFIERS.md`.
+Структура презентации: `PRESENTATION_OUTLINE.md`.
+
+---
+
+## 2. Требования к окружению
+
+- Python 3.11 или новее (проверено на 3.14).
+- Linux, macOS или Windows с WSL.
+- Свободное место: около 1 ГБ с учётом виртуального окружения и ресурсов
+  NLTK.
+- Интернет — только для установки зависимостей и скачивания ресурсов NLTK.
+  После этого эксперименты выполняются локально.
+
+Проверьте версию:
+
+```bash
+python --version
+```
+
+Если версия ниже 3.11, обновите Python: в коде используется стандартный
+модуль `tomllib`.
+
+---
+
+## 3. Первичная установка
+
+### 3.1. Перейти в каталог лабораторной
+
+```bash
+cd nlp/lab1
+```
+
+Все дальнейшие команды выполняются из этого каталога.
+
+### 3.2. Создать виртуальное окружение
+
+Можно создать отдельное окружение внутри `lab1`:
+
+```bash
+python -m venv .venv
+source .venv/bin/activate
+```
+
+На Windows:
+
+```powershell
+python -m venv .venv
+.venv\Scripts\activate
+```
+
+Если в корне проекта уже есть окружение `.venv`, можно использовать его:
+
+```bash
+source ../.venv/bin/activate
+```
+
+### 3.3. Установить зависимости
+
+```bash
+python -m pip install --upgrade pip
+python -m pip install -r requirements.txt
+```
+
+Проверьте, что установились основные библиотеки:
+
+```bash
+python -c "import numpy, pandas, scipy, sklearn, nltk, matplotlib; print('ok')"
+```
+
+### 3.4. Скачать ресурсы NLTK
+
+```bash
+python scripts/download_nltk_data.py
+```
+
+Ресурсы сохраняются в `lab1/nltk_data`:
+
+- `punkt`, `punkt_tab` — токенизация и разбиение на предложения;
+- `averaged_perceptron_tagger`, `averaged_perceptron_tagger_eng` —
+  частеречная разметка;
+- `stopwords` — стоп-слова для content-LSA.
+
+Без этих ресурсов POS-признаки отключатся, и результаты будут неполными.
+
+### 3.5. Проверить окружение
+
+```bash
+python scripts/check_environment.py
+```
+
+Скрипт печатает версии библиотек, SHA-256 корпуса, число реплик и классов,
+проверяет токенизатор и POS-теггер. В конце должно быть:
+
+```text
+Окружение готово.
+```
+
+Если POS-теггер недоступен, повторите шаг 3.4.
+
+---
+
+## 4. Проверка перед полной серией
+
+### 4.1. Тесты
+
+```bash
+python -m pytest -q
+```
+
+Ожидается:
+
+```text
+33 passed
+```
+
+Если тесты не проходят, не запускайте полную серию: сначала пришлите текст
+ошибки.
+
+### 4.2. Быстрый smoke-test
+
+```bash
+python -m stylometry.cli audit
+python -m stylometry.cli features
+python -m stylometry.cli run --preset length --model logreg --folds 3 --repeats 1
+```
+
+Проверьте, что появились:
+
+- `data/processed/audit.md`;
+- `data/processed/lines.csv`;
+- `features/stylometric.csv`;
+- `features/feature_dictionary.csv`;
+- каталог `results/length_logreg_<timestamp>/` с `summary.md`.
+
+Если всё создалось, окружение готово.
+
+---
+
+## 5. Полная серия экспериментов
+
+### 5.1. Основной запуск
+
+Из каталога `lab1`:
+
+```bash
+bash scripts/run_all_experiments.sh
+```
+
+Если используется окружение в корне проекта:
+
+```bash
+PYTHON=../.venv/bin/python bash scripts/run_all_experiments.sh
+```
+
+Скрипт последовательно запускает все стадии, пишет общий лог в
+`results/logs/run_all_<timestamp>.log` и останавливается при первой ошибке.
+
+### 5.2. Быстрый режим
+
+Для проверки, что всё работает, но без финальных чисел:
+
+```bash
+QUICK=1 bash scripts/run_all_experiments.sh
+```
+
+В быстром режиме:
+
+- все повторы равны 1;
+- вложенная кросс-валидация пропускается;
+- полная серия занимает меньше времени.
+
+Результаты быстрого режима **нельзя** использовать в отчёте как финальные.
+
+### 5.3. Настройка повторов
+
+По умолчанию:
+
+| Переменная | Значение | Что регулирует |
+|---|---:|---|
+| `REPEATS_MAIN` | 5 | основные модели |
+| `REPEATS_FAST` | 3 | baseline, benchmark, delta |
+| `REPEATS_LSA` | 3 | отдельные варианты LSA |
+| `REPEATS_ABLATE` | 3 | поочерёдное исключение групп |
+| `REPEATS_DELTA` | 2 | дельта Барроуза |
+| `SKIP_NESTED` | 0 | 1 — пропустить вложенную CV |
+
+Пример:
+
+```bash
+REPEATS_MAIN=5 REPEATS_FAST=3 SKIP_NESTED=1 \
+    bash scripts/run_all_experiments.sh
+```
+
+### 5.4. Долгий запуск через tmux или nohup
+
+Полная серия может занимать несколько часов. Если подключение к машине
+может прерваться, используйте `tmux`:
+
+```bash
+tmux new -s nlp
+PYTHON=../.venv/bin/python bash scripts/run_all_experiments.sh
+# Ctrl+B, затем D — выйти из tmux, процесс продолжит работу
+```
+
+Вернуться:
+
+```bash
+tmux attach -t nlp
+```
+
+Или `nohup`:
+
+```bash
+nohup env PYTHON=../.venv/bin/python \
+    bash scripts/run_all_experiments.sh >/dev/null 2>&1 &
+```
+
+Следить за логом:
+
+```bash
+tail -f results/logs/run_all_*.log
+```
+
+### 5.5. Ориентировочное время
+
+Время зависит от процессора, числа потоков и версии библиотек. Ориентиры для
+обычного ноутбука:
+
+| Стадия | Время |
+|---|---:|
+| Проверка окружения и тесты | 1–2 мин |
+| Аудит и ручные признаки | 1–2 мин |
+| Нижние границы | 5–10 мин |
+| Классическая стилометрия | 3–6 мин |
+| Ручные признаки + n-граммы | 4–8 мин |
+| Основной объединённый набор | 5–10 мин |
+| Варианты LSA | 10–20 мин |
+| Перебор числа компонент LSA | 5–10 мин |
+| Сравнение классификаторов | 15–30 мин |
+| Поочерёдное исключение групп | 20–40 мин |
+| Дельта Барроуза | 3–5 мин |
+| Верификация и кластеризация | 3–5 мин |
+| Вложенная кросс-валидация | 15–40 мин |
+| Анализ длины и подгрупп | 1–2 мин |
+
+Итого: ориентировочно **1,5–3 часа**, в зависимости от машины. Если
+включены все повторы и вложенная CV, может быть дольше.
+
+---
+
+## 6. Что делает каждая стадия
+
+### 6.1. Аудит корпуса (E0)
+
+```bash
+python -m stylometry.cli audit
+```
+
+Создаёт:
+
+- `data/processed/audit.md` — отчёт по корпусу;
+- `data/processed/lines.csv` — нормализованные реплики;
+- `data/processed/figures/class_distribution.png`.
+
+Проверьте: 2005 реплик, 34 персонажа, нет точных дублей, распределение
+классов совпадает с `stats_per_character.xlsx`.
+
+### 6.2. Ручные признаки
+
+```bash
+python -m stylometry.cli features
+```
+
+Создаёт:
+
+- `features/stylometric.csv` — матрица 2005 × 454;
+- `features/feature_dictionary.csv` — описание признаков.
+
+### 6.3. Нижние границы (E1)
+
+Запускаются:
+
+- `length` + Dummy — частотная базовая модель;
+- `length` + логистическая регрессия;
+- `char` + MultinomialNB;
+- `pos` + MultinomialNB;
+- `fw` + логистическая регрессия.
+
+Смысл: понять, сколько сигнала даёт только длина, только символьные
+n-граммы, только POS и только функциональные слова.
+
+### 6.4. Классическая стилометрия (E2)
+
+```bash
+python -m stylometry.cli run --preset stylometry --model logreg --repeats 5
+python -m stylometry.cli run --preset stylometry --model linear_svc --repeats 5
+```
+
+Проверяет гипотезу H1 на ручных признаках без n-грамм и LSA.
+
+### 6.5. Ручные признаки + n-граммы (E4)
+
+```bash
+python -m stylometry.cli run --preset stylometry_pos --model logreg --repeats 5
+```
+
+Показывает вклад символьных и частеречных n-грамм.
+
+### 6.6. Основной объединённый набор (E8)
+
+```bash
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --repeats 5
+```
+
+Основной результат: ручные признаки + char/POS n-граммы + LSA по словам и
+POS.
+
+### 6.7. Варианты LSA (E5, E7)
+
+Запускаются отдельно:
+
+```bash
+python -m stylometry.cli run --preset lsa_word    --model logreg --repeats 3
+python -m stylometry.cli run --preset lsa_word2   --model logreg --repeats 3
+python -m stylometry.cli run --preset lsa_content --model logreg --repeats 3
+python -m stylometry.cli run --preset lsa_fw      --model logreg --repeats 3
+python -m stylometry.cli run --preset lsa_char    --model logreg --repeats 3
+python -m stylometry.cli run --preset lsa_pos     --model logreg --repeats 3
+```
+
+Сравнение `lsa_word`/`lsa_content` с `lsa_pos`/`lsa_fw`/`lsa_char` — это
+основной эксперимент «стиль или тема».
+
+### 6.8. Перебор числа компонент LSA (E5)
+
+```bash
+python -m stylometry.cli lsa-sweep \
+    --variants word,content,fw,char,pos \
+    --components 50,100,200 \
+    --model logreg --repeats 1
+```
+
+Результат: `results/lsa_sweep_<timestamp>/lsa_sweep.csv` с macro-F1 для
+каждой пары «вариант — число компонент».
+
+### 6.9. Сравнение классификаторов (E6)
+
+```bash
+python -m stylometry.cli benchmark \
+    --preset stylometry_pos_lsa \
+    --models logreg,linear_svc,random_forest,multinomial_nb,knn \
+    --repeats 3
+```
+
+Результат: `results/benchmark_<preset>_<timestamp>/summary.csv` и отдельные
+каталоги по моделям.
+
+### 6.10. Поочерёдное исключение групп признаков (E3)
+
+```bash
+python -m stylometry.cli ablate \
+    --preset stylometry_pos_lsa --model logreg --repeats 3
+```
+
+Результат: `results/ablation_<preset>_<timestamp>/ablation.csv` с колонкой
+`delta_macro_f1`. Отрицательное значение означает, что удаление группы
+ухудшило качество.
+
+### 6.11. Дельта Барроуза (E11)
+
+```bash
+python -m stylometry.cli delta \
+    --features 100,200,500 --metrics manhattan,cosine --repeats 2
+```
+
+Классическая стилометрическая базовая модель; результат —
+`results/delta_<timestamp>/delta_sweep.csv`.
+
+### 6.12. Верификация «свой/чужой» (E12)
+
+```bash
+python -m stylometry.cli verify --preset stylometry_pos_lsa
+```
+
+Результат: `results/verification_<preset>_<timestamp>/summary.json` с
+ROC-AUC, EER, TAR@FAR=1% и top-1.
+
+### 6.13. Кластеризация (E13)
+
+```bash
+python -m stylometry.cli cluster --preset stylometry_pos_lsa
+```
+
+Результат: `results/clustering_<preset>_<timestamp>/clustering_metrics.csv`
+и `character_centroids.csv`.
+
+### 6.14. Вложенная кросс-валидация (E16)
+
+```bash
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --nested
+python -m stylometry.cli run --preset stylometry_pos_lsa --model linear_svc --nested
+```
+
+Внешние блоки — оценка качества, внутренние — настройка гиперпараметров.
+Сохраняются `best_params.json` и метрики внешних блоков.
+
+### 6.15. Анализ длины и подгрупп (E9, E14, E15)
+
+После основного запуска:
+
+```bash
+python -m stylometry.cli length-report \
+    --run-dir results/<каталог_основного_запуска>
+
+python -m stylometry.cli subgroup-report \
+    --run-dir results/<каталог_основного_запуска> --group gender
+
+python -m stylometry.cli subgroup-report \
+    --run-dir results/<каталог_основного_запуска> --group age_group
+
+python -m stylometry.cli subgroup-report \
+    --run-dir results/<каталог_основного_запуска> --group species
+
+python -m stylometry.cli compare \
+    --run-a results/<каталог_основного_запуска> \
+    --run-b results/<каталог_stylometry_pos> \
+    --out results/<каталог_основного_запуска>/comparison_main_vs_pos.json
+```
+
+`compare` корректно работает, если запуски сделаны с одинаковыми `--folds` и
+`--seed`.
+
+---
+
+## 7. Если серия прервалась
+
+Скрипт останавливается при первой ошибке. Уже созданные каталоги результатов
+не удаляются. Порядок действий:
+
+1. Посмотрите конец лога:
+
+```bash
+tail -n 100 results/logs/run_all_*.log
+```
+
+2. Исправьте причину (см. раздел 10).
+3. Запустите пропущенные команды вручную из раздела 6. Можно продолжить с
+   нужной стадии, не повторяя всю серию.
+4. Новые результаты появятся в отдельных каталогах с новыми метками времени.
+
+Не удаляйте старые каталоги: по ним можно сравнить, что изменилось.
+
+---
+
+## 8. Упаковка результатов
+
+После завершения серии:
+
+```bash
+bash scripts/collect_results.sh
+```
+
+Скрипт создаёт:
+
+```text
+exports/
+├── nlp_lab1_results_<timestamp>.tar.gz
+├── nlp_lab1_results_<timestamp>.tar.gz.sha256
+└── nlp_lab1_results_<timestamp>_manifest.txt
+```
+
+В архив входят:
+
+- `results/` — все запуски, метрики, предсказания, матрицы ошибок, графики,
+  логи;
+- `features/` — матрицы признаков и LSA;
+- `data/processed/` — очищенные реплики и аудит;
+- `data/metadata/` — метаданные персонажей;
+- манифест с версиями библиотек, SHA-256 корпуса и списком файлов.
+
+Передайте коллеге **архив** и **файл `.sha256`**. Если архив большой,
+можно передать его через файлообменник, но контрольную сумму нужно сохранить.
+
+Проверка архива на стороне получателя:
+
+```bash
+sha256sum -c nlp_lab1_results_<timestamp>.tar.gz.sha256
+tar -xzf nlp_lab1_results_<timestamp>.tar.gz
+```
+
+---
+
+## 9. Что проверить перед отправкой
+
+- [ ] `python scripts/check_environment.py` завершился без ошибок.
+- [ ] `python -m pytest -q` — все тесты пройдены.
+- [ ] В логе нет строк `ERROR`, `Traceback`, `FAILED`.
+- [ ] Основной запуск `stylometry_pos_lsa` содержит `summary.md`,
+      `pooled_metrics.csv`, `per_class_report.csv`, `confusion_matrix.csv`.
+- [ ] Есть запуски `stylometry`, `stylometry_pos`, `stylometry_pos_lsa`.
+- [ ] Есть `results/benchmark_*`, `results/ablation_*`, `results/lsa_sweep_*`.
+- [ ] Есть `results/verification_*` и `results/clustering_*`.
+- [ ] Есть `results/logs/run_all_*.log`.
+- [ ] `collect_results.sh` создал архив и `.sha256`.
+
+---
+
+## 10. Типичные проблемы
+
+### 10.1. `ModuleNotFoundError`
+
+Не установлены зависимости или выбран другой интерпретатор.
+
+```bash
+python -m pip install -r requirements.txt
+python scripts/check_environment.py
+```
+
+Если используете PyCharm, проверьте, что выбран интерпретатор
+`nlp/.venv/bin/python` или `lab1/.venv/bin/python`.
+
+### 10.2. Ошибка NLTK: `Resource punkt_tab not found`
+
+```bash
+python scripts/download_nltk_data.py
+```
+
+Если интернет недоступен, скопируйте каталог `nltk_data` с другой машины в
+`lab1/nltk_data`.
+
+### 10.3. `UnicodeDecodeError` при чтении корпуса
+
+Корпус хранится в cp1251. Код сам пробует cp1251, UTF-8 и koi8-r. Если вы
+пересохранили CSV в другой кодировке, верните cp1251 или используйте
+`dialogues.xlsx`.
+
+### 10.4. `ValueError: empty vocabulary`
+
+Обычно означает, что в обучающем блоке не осталось признаков. Проверьте:
+
+- не пустой ли корпус;
+- не отключены ли все блоки признаков;
+- не выбран ли `multinomial_nb` с одним LSA без char/POS.
+
+Для стандартных пресетов этой ошибки быть не должно.
+
+### 10.5. Не хватает памяти
+
+Уменьшите размерности:
+
+```bash
+python -m stylometry.cli run --preset stylometry_pos --model logreg --repeats 3
+```
+
+Или закройте другие приложения. Основной объединённый набор с LSA требует
+больше памяти, чем `stylometry_pos`.
+
+### 10.6. Серия идёт слишком долго
+
+- Используйте `QUICK=1` для проверки.
+- Уменьшите `REPEATS_FAST`, `REPEATS_ABLATE`, `REPEATS_LSA`.
+- Пропустите вложенную CV: `SKIP_NESTED=1`.
+- Для финального отчёта всё равно нужны основные повторы; не заменяйте их
+  быстрым режимом.
+
+### 10.7. `tomllib` не найден
+
+Нужен Python 3.11+. Обновите интерпретатор или создайте окружение на 3.11+.
+
+---
+
+## 11. Полная матрица экспериментов
+
+| ID | Команда | Что проверяет |
+|---|---|---|
+| E0 | `audit` | Корпус, классы, длины, дубли |
+| E1 | `run --preset length/char/pos/fw` | Нижние границы |
+| E2 | `run --preset stylometry` | Ручная стилометрия |
+| E3 | `ablate --preset stylometry_pos_lsa` | Вклад групп признаков |
+| E4 | `run --preset stylometry_pos` | Вклад char/POS n-грамм |
+| E5 | `lsa-sweep` и `run --preset lsa_*` | Варианты и число компонент LSA |
+| E6 | `benchmark --preset stylometry_pos_lsa` | Сравнение классификаторов |
+| E7 | `lsa-sweep` | Стиль или тема |
+| E8 | `run --preset stylometry_pos_lsa` | Объединение блоков |
+| E9 | `length-report` | Влияние длины |
+| E10 | `benchmark` + per-class отчёт | Дисбаланс классов |
+| E11 | `delta` | Дельта Барроуза |
+| E12 | `verify` | Верификация «свой/чужой» |
+| E13 | `cluster` | Кластеризация персонажей |
+| E14 | `subgroup-report` | Пол, возраст, вид |
+| E15 | `compare` | Мак-Немар и бутстрэп-разница |
+| E16 | `run --nested` | Вложенная кросс-валидация |
+| E17 | не реализовано | Агрегация в псевдодокументы по 30 предложений |
+
+Подробные команды и интерпретация — в `EXPERIMENTS.md`.
+
+---
+
+## 12. Что именно отправлять обратно
+
+Минимальный комплект:
+
+1. `exports/nlp_lab1_results_<timestamp>.tar.gz`;
+2. `exports/nlp_lab1_results_<timestamp>.tar.gz.sha256`;
+3. краткий ответ: на какой машине запускали, сколько времени заняла серия,
+   какие стадии завершились, какие ошибки были.
+
+Если серия не завершилась, отправьте:
+
+- лог `results/logs/run_all_*.log`;
+- архив с уже готовыми результатами;
+- текст ошибки и команду, на которой она возникла.
+
+---
+
+## 13. Что нельзя менять без согласования
+
+Чтобы результаты оставались сопоставимыми:
+
+- не меняйте корпус и разметку классов;
+- не удаляйте реплики и не переименовывайте персонажей;
+- не меняйте `--folds`, `--seed` и схему стратификации в основных запусках;
+- не меняйте определения метрик;
+- не настраивайте гиперпараметры на тестовых блоках;
+- не выбирайте признаки по тестовой выборке;
+- не заменяйте macro-F1 на долю правильных ответов как основную метрику.
+
+Если хотите добавить новый эксперимент, создайте отдельный сценарий и
+сохраните его в `experiments/`, не изменяя основные модули.
+
+---
+
+## 14. Краткая шпаргалка
+
+```bash
+cd nlp/lab1
+source ../.venv/bin/activate
+python scripts/check_environment.py
+python -m pytest -q
+
+# быстрая проверка
+QUICK=1 bash scripts/run_all_experiments.sh
+
+# полная серия
+bash scripts/run_all_experiments.sh
+
+# упаковка результатов
+bash scripts/collect_results.sh
+
+# посмотреть лог
+tail -f results/logs/run_all_*.log
+```
+
+Если что-то пошло не так, сначала пришлите лог и текст ошибки, а не только
+вывод последней команды.
diff --git a/lab1/configs/default.toml b/lab1/configs/default.toml
new file mode 100644
index 0000000..70b80a9
--- /dev/null
+++ b/lab1/configs/default.toml
@@ -0,0 +1,27 @@
+# Конфигурация по умолчанию для экспериментов.
+# Все пути указаны относительно каталога lab1.
+
+[data]
+path = "data/dialogues.csv"
+cache = "data/processed/tokenized.pkl"
+
+[cv]
+folds = 5
+repeats = 5
+seed = 42
+
+[experiment]
+# Основной набор: ручные признаки + символьные и частеречные n-граммы + LSA.
+preset = "stylometry_pos_lsa"
+model = "logreg"
+
+[features]
+# Эти параметры описаны в stylometry/pipelines.py и могут переопределяться
+# программно. В командной строке удобнее выбирать готовый preset.
+# Параметры, которые зависят от preset (например, lsa_variants и
+# stylo_normalize), здесь не задаются, чтобы не переопределять preset.
+char_ngram_range = [2, 5]
+pos_ngram_range = [1, 3]
+lsa_components = 100
+min_df = 3
+stylo_scaler = "standard"
diff --git a/lab1/data/corpus_report_after.txt b/lab1/data/corpus_report_after.txt
new file mode 100644
index 0000000..686ef9a
--- /dev/null
+++ b/lab1/data/corpus_report_after.txt
@@ -0,0 +1,49 @@
+=== СТАТИСТИКА ПО ПЕРСОНАЖАМ ===
+
+           реплик  средняя_длина_символов  средняя_длина_слов  мин_символов  макс_символов  мин_слов  макс_слов
+Персонаж                                                                                                       
+Abigail       139                    94.7                18.0            17            260         4         51
+Alex          120                    83.5                16.1            14            246         3         48
+Haley         112                    86.4                16.8            21            225         4         43
+Sebastian     106                    92.6                17.6            31            248         5         47
+Emily          97                   103.6                19.7            28            214         5         42
+Maru           96                    90.9                16.9            25            355         4         61
+Sam            94                   108.3                20.8            22            265         4         48
+Elliott        91                    88.4                16.7            23            206         4         44
+Leah           89                   101.7                19.3            31            306         5         56
+Penny          84                   112.1                21.3            26            336         6         60
+Lewis          67                   119.9                22.0            37            274         4         49
+Pam            65                    98.6                19.4            30            270         6         53
+Harvey         62                   106.2                19.6            35            245         5         52
+Shane          62                    88.0                17.1            22            204         4         40
+Demetrius      56                   110.0                19.6            29            214         5         38
+Robin          54                   111.8                20.7            46            299         9         53
+Pierre         53                   114.4                21.5            45            286         7         52
+Linus          51                   103.7                20.1            35            254         7         51
+Jodi           48                   118.5                23.1            33            274         8         55
+Evelyn         47                   108.0                20.8            37            325         7         63
+Willy          42                   114.0                21.8            16            333         3         61
+Clint          39                   115.1                21.7            31            247         6         51
+Gus            37                   105.8                19.5            36            228         6         43
+Jas            34                    60.7                11.9            13            123         4         26
+George         32                   108.4                20.5            34            299         6         56
+Marnie         31                   110.0                21.1            58            274        11         48
+Kent           30                    99.3                19.1            44            406         8         80
+Leo            30                   120.1                23.5            26            278         4         53
+Krobus         28                    91.5                17.2            38            239         6         40
+Vincent        26                    54.8                10.7            17            138         3         27
+Wizard         24                   119.7                21.8            50            206         9         35
+Dwarf          23                   100.4                17.8            25            208         4         36
+Sandy          21                   139.3                26.3            58            282         9         60
+Mister Qi      15                   114.5                21.0            33            183         6         35
+
+=== ОБЩАЯ СТАТИСТИКА ===
+
+Всего реплик: 2005
+Персонажей: 34
+Средняя длина (симв.): 100.5
+Средняя длина (слов):  19.1
+Медиана (симв.):       91.0
+Медиана (слов):        17.0
+Мин/макс (симв.):      13 / 406
+Мин/макс (слов):       3 / 80
diff --git a/lab1/data/corpus_report_before_cleaning.txt b/lab1/data/corpus_report_before_cleaning.txt
new file mode 100644
index 0000000..9f33463
--- /dev/null
+++ b/lab1/data/corpus_report_before_cleaning.txt
@@ -0,0 +1,158 @@
+=== ОТЧЁТ ПО КОРПУСУ ===
+
+Всего реплик: 2198
+Уникальных реплик: 2177
+Персонажей: 36
+
+--- Реплик у каждого персонажа ---
+  Abigail         148
+  Haley           127
+  Sebastian       124
+  Alex            123
+  Sam             115
+  Maru            112
+  Penny           105
+  Emily           103
+  Leah            96
+  Elliott         95
+  Lewis           71
+  Harvey          70
+  Shane           69
+  Pam             69
+  Demetrius       59
+  Pierre          58
+  Robin           58
+  Linus           53
+  Jodi            53
+  Evelyn          49
+  Clint           44
+  Willy           42
+  Jas             40
+  Marnie          39
+  Gus             38
+  Kent            34
+  George          33
+  Krobus          28
+  Vincent         26
+  Wizard          24
+  Dwarf           23
+  Sandy           21
+  LeoMainland     20
+  Mister Qi       15
+  Leo             12
+  Gil             2
+
+--- Дубли внутри персонажа: 13 ---
+  Elliott         x2  "Agh, it's still wriggling! Get that abomination away from me!"
+  Haley           x2  "Aw, thanks!"
+  Haley           x2  "Is that so...?"
+  Haley           x2  "Oh! I thought you knew. This room's for developing photos."
+  Lewis           x2  "Everyone seems a little happier on the weekends, don't you think? I'm on the job seven days a week, "
+  Lewis           x2  "I hope things are going well for you on the farm."
+  Penny           x2  "I see."
+  Pierre          x2  "Hello, er... ?"
+  Robin           x2  "Our little plan worked out well, don't you think? Pam and Penny seem really happy."
+  Sam             x4  "Hey, you know what? That's exactly the kind of style I've been thinking about... for the band."
+  Sam             x2  "Oh, hi! It's good to see you again. Take it easy."
+  Sam             x2  "Well, my Dad is back. Have you met him? I'm just glad he's okay."
+  Sam             x2  "Yeah... That's what happened!"
+
+--- Дубли между персонажами: 6 ---
+  "Are you having fun, <PLAYER>? You need to remember to take breaks now and then too!"
+    -> Demetrius, Emily
+  "Don't you have work to do?"
+    -> Clint, Shane
+  "Thank you!"
+    -> Emily, Evelyn
+  "That was fun!"
+    -> Maru, Robin
+  "Wow, tastes just like golden walnuts! This is great!"
+    -> Leo, LeoMainland
+  "sigh ..."
+    -> Demetrius, Haley
+
+--- Короткие реплики (<=3 слов): 84 ---
+  Abigail         "Oh, hi."
+  Abigail         "...You're cute."
+  Abigail         "Hmm, interesting..."
+  Abigail         "I see..."
+  Abigail         "...That's true."
+  Abigail         "<PLAYER>..."
+  Abigail         "<PLAYER>... I'm alright."
+  Abigail         "...okay."
+  Abigail         "...Um, okay. Sorry."
+  Alex            "Finally, we're here!"
+  Alex            "Uh... never mind."
+  Alex            "Hey, nice tan."
+  Alex            "What's up, <PLAYER>?"
+  Alex            "<PLAYER>! Um... Hi."
+  Clint           "I'm hungry..."
+  Clint           "How'd you know?"
+  Clint           "Hahahaha. Good one."
+  Demetrius       "sigh ..."
+  Elliott         "Hmmph...forget it."
+  Emily           "Thank you!"
+  Emily           "Hi <PLAYER>!"
+  Emily           "Hi! ...Yes?"
+  Evelyn          "Thank you!"
+  George          "...Hmmph."
+  Gil             "... snore ..."
+  Haley           "Ew... No."
+  Haley           "Nice shoes."
+  Haley           "sigh ..."
+  Haley           "Aw, thanks!"
+  Haley           "Is that so...?"
+  Haley           "Aw, thanks!"
+  Haley           "Is that so...?"
+  Haley           "It's fall already?"
+  Harvey          "Sigh ... rude."
+  Harvey          "...You're right. ...Thanks."
+  Harvey          "sigh"
+  Jas             "... Hi..."
+  Jas             "I'm scared..."
+  Jas             "Ew, gross!"
+  Jas             "Hi, Mr. <PLAYER>."
+  Jas             "What? giggle ...hi."
+  Jas             "...Hi."
+  Jodi            "What a treat!"
+  Jodi            "Hi. Need something?"
+  Leah            "You pig!"
+  Leo             "Hi... Everything okay?"
+  Lewis           "Hi, Gus!"
+  Linus           "Find anything good?"
+  Marnie          "I feel warm."
+  Marnie          "Phew... It's hot!"
+  Maru            "That was fun!"
+  Maru            "... <PLAYER>..."
+  Maru            "See you later."
+  Pam             "Hi, stranger."
+  Penny           "Oh... No, thanks."
+  Penny           "I see."
+  Penny           "Oh... Uh, sure."
+  Penny           "Oh...I'm sorry."
+  Penny           "That's right..."
+  Penny           "No!"
+  Penny           "Y...You don't? ...!!!"
+  Penny           "Me too."
+  Penny           "...sorry."
+  Penny           "I see."
+  Penny           "...yes?"
+  Pierre          "Hello, er... ?"
+  Pierre          "Hello, er... ?"
+  Pierre          "Hey, good luck."
+  Robin           "That was fun!"
+  Sam             "Hey, farmer."
+  Sam             "Oh, I see."
+  Sam             "What!? lying!"
+  Sebastian       "Hey..."
+  Sebastian       "Um... need something?"
+  Sebastian       "Okay, that'll work."
+  Sebastian       "Great."
+  Sebastian       "Huh. Sounds boring."
+  Shane           "...No."
+  Shane           "Okay, Dad..."
+  Shane           "Oh, wow..."
+  Shane           "...Yeah, me too."
+  Vincent         "Hi there, mister!"
+  Vincent         "Hi Uncle <PLAYER>!"
+  Willy           "Ahoy there, son."
diff --git a/lab1/data/dialogues.csv b/lab1/data/dialogues.csv
new file mode 100644
index 0000000..4963224
--- /dev/null
+++ b/lab1/data/dialogues.csv
@@ -0,0 +1,2006 @@
+�������;��������
+Oh, that's right... I heard someone new was moving onto that old farm. It's kind of a shame, really. I always enjoyed exploring those overgrown fields by myself.;Abigail
+Hey, sure. Let's do it. We can be embarrassed together.;Abigail
+Hey, you remembered my birthday... Thanks, <PLAYER>! This is really nice!;Abigail
+Hey, you remembered my birthday... Thanks, <PLAYER>! Oh, it's, uh... sniff ... what is it?;Abigail
+Wow... the color is so beautiful, like an amethyst. Thanks!;Abigail
+Disgusting... I love it!;Abigail
+Hey, this is my favorite magazine! My Dad won't let me have a subscription, so I have to get it on the black market like this...;Abigail
+Whoa, that's a creepy looking book... I love it! Flip flip flip ... Ooh... It's full of monsters... I'll have to study this before going into the caves... Thanks!;Abigail
+Whoa... Where'd you find this? The workmanship is fascinating... It's a little too rusty to use, but I'll add it to my collection.;Abigail
+This makes you wonder... Thousands of years ago, there was probably a little purple-haired cavewoman who played this flute, just like me. ...I'll have to wash it before pressing it to my lips, though.;Abigail
+Wow... it's just my color! Thanks, <PLAYER>! Thank you! This is my favorite stone. It's so pretty.;Abigail
+I want to go exploring, but I can tell my Mom is kind of worried. I don't want to add to her stress.;Abigail
+And just like that, everything's back to normal. It's like magic!;Abigail
+Have you seen all the crazy trees around town? Everything looks so different... I feel like we're in a fairy world.;Abigail
+Don't tell anyone, but I snuck into the old community center last night... I think it's haunted!;Abigail
+Oh, wow... <PLAYER>! Are you sure you want to give this to me? It's so rare! I'll admit, my mouth is watering already...;Abigail
+I heard you've been exploring the old mines. That sounds like fun! ...I'm working up the courage to go there myself...;Abigail
+Sam and Sebastian asked me to play drums in their band... I've never played drums before, but it's actually pretty fun. You have to learn how to move your limbs to different rhythms, at the same exact time. I feel like my brain is growing.;Abigail
+Hey, I finally beat the first level of Journey Of The Prairie King! Watching you play helped a lot.;Abigail
+Sebastian mentioned that he heard strange music while having dinner the other night... I didn't tell him it was us! It seemed better to keep it a mystery.;Abigail
+The spirits have a mind of their own...;Abigail
+Feeling pretty good about the egg hunt, huh? Well, next year you won't be so lucky...;Abigail
+I'm still mad that you beat me at the egg hunt last year. You'd better start practicing...;Abigail
+Uh... <PLAYER>? We've been dating for almost a year. Is this going anywhere?;Abigail
+I heard you and Sebastian started dating... I didn't realize you guys had a thing...;Abigail
+Hey, I heard you and Sam got together... He's a really good guy! We've been friends for a long time. You'd better be good to him, or you'll see my bad side!;Abigail
+Congratulations on the marriage. I guess this means Sebastian won't be moving to the city like he always talked about. That'll be nice. I'd be kind of sad if I never saw him again. We've always been good friends.;Abigail
+Haha... I can't believe Sam is a married man, now. Who would've thought? Joking aside, I think he'll make a great father some day.;Abigail
+Did you hear about the new movie theater in town? I'm excited to check it out!;Abigail
+I hope a mermaid visits the beach.;Abigail
+I'm gonna ask Willy if we can take the scenic route back.;Abigail
+Maybe we'll see a sea monster on the way back.;Abigail
+The beach is okay, but it gets boring pretty quick. I'd rather check out that dark jungle, or the volcano!;Abigail
+Hmm, you think I could sneak my guinea pig onto the island? Nah, David's probably not much of a swimmer anyways.;Abigail
+My skin gets burnt really easily. So I'm just gonna stay under this umbrella for a while. I like it in the shade. Too much sun makes me dizzy.;Abigail
+Hey <PLAYER>, want to build a sand castle with me? Actually, a sand dungeon might be cooler!;Abigail
+Hey <PLAYER>, you think you could build a raft? I saw a few cool islands on the way here I want to visit.;Abigail
+An island this big has got to be hiding some buried treasure. Quick, get me a shovel! Relax, I'm just kidding.;Abigail
+Gus made me a dark purple refresher, to match my hair. It tastes great!;Abigail
+Bummer... Willy says we have to stay on the beach... And I brought my sword and everything...;Abigail
+B... But... I wanted to dance with... Just kidding! Of course I'm dancing with you. I love you.;Abigail
+That's flattering... but I'm gonna have to say no. Sorry.;Abigail
+What are you doing here? Just leave me alone!;Abigail
+Don't talk to me!;Abigail
+sigh ... I can't stay mad at you anymore, <PLAYER>. Let's just move on with our lives.;Abigail
+Oh... I... I understand. You see me as just a friend... I guess we weren't really a good match...;Abigail
+Oh, hey. Taking a break from work?;Abigail
+Oh, hi! Do you ever hang out at the cemetery? It's a peaceful place to spend some time alone.;Abigail
+Ugh... I'm not in a good mood right now. I said I'm not in a good mood... what do you want?;Abigail
+... sigh ... I know my parents mean well, but sometimes they just cannot understand my point of view. Weren't they ever young?;Abigail
+Hey. Sorry in advance if I say anything rude. I didn't get much sleep last night. What do you want?;Abigail
+Oh no, I think my Dad's going to cook dinner tonight... I don't feel like doing anything today...;Abigail
+Wow, I just realized it's Friday. Sometimes I totally lose track of time.;Abigail
+We're all going to hang out in the saloon this evening. It's kind of like a Friday tradition, I guess.;Abigail
+The fresh mountain air is nice on a day like this. I wonder if the frogs will make an appearance soon.;Abigail
+Oh man... I've been pushing off my homework all weekend. Looks like I'll be pulling another all-nighter...;Abigail
+I wonder what would happen if I spent all night in the graveyard? <PLAYER>, what do you think happens to us after we die? I have no idea. We come back as spooky ghosts. We go to Heaven. Our energy bodies enter the astral plane. Nothing. We just cease to exist.;Abigail
+I guess you think nothing would happen, right? Maybe a wicked ghost would appear!;Abigail
+Well that's an interesting theory!;Abigail
+You know, that wouldn't be so bad.;Abigail
+Oh, hi <PLAYER>. Taking a break from your work? Me too. Oh! Nothing physical... just some online classes I'm taking.;Abigail
+Hi, I'm glad to see you. I want to take my mind off things for a while... how is your day going?;Abigail
+You came all this way to visit me? That's nice. So have you been exploring the mountain caves at all? Interesting. I'd like to go there myself one of these days.;Abigail
+I remember when you first arrived in town, I was a little sad that those old woods would be turned into farmland. But now I'm really glad you moved here! ...Oh, <PLAYER>! Hi. Want to hang out for a while? Here! Let me read your palm.;Abigail
+Hi. Do you feel like everything seems unreal lately? Not in a bad way, though.;Abigail
+The birds were happily chirping today. They're too simple to understand the future. It's better to be like them, isn't it?;Abigail
+Great! I'm sure you'll be able to beat this level with ease.;Abigail
+Oh? sigh ... I guess I'll never get past this level, then...;Abigail
+See? You understand why I'm out here, then. I've lived in the valley my whole life, but I've never really done anything memorable. I want to go on an adventure!;Abigail
+Sure. I don't want to kill anything, either. I just want to do a little exploring without risking my neck! I've lived in the valley my whole life, but I've never really done anything memorable. I want to go on an adventure!;Abigail
+Oh, it's because I'm a girl... isn't it? Ugh...;Abigail
+Really? That's surprising. I've lived in the valley my whole life, but I've never really done anything memorable. I want to go on an adventure!;Abigail
+Yea... I guess you would be one to appreciate this somber ambience.;Abigail
+Yeah... I guess my own life is okay...;Abigail
+I would feel better if I could believe that... But I don't.;Abigail
+I guess when you have something to lose, it's normal to be afraid...;Abigail
+Summer's here. It's not really my favorite season. The air is so muggy I feel like I'm floating in soup.;Abigail
+My pet guinea pig, David, just hates this hot weather. He's fussy.;Abigail
+The air is so thick with honey and nectar all summer. I almost feel dizzy.;Abigail
+Have you decorated your house at all? That's what I'd be doing if I had a house.;Abigail
+I went to the beach last night, after dark... Sometimes you can see strange lights bobbing over the sea. Or maybe that was just a dream I had... It doesn't feel real anymore.;Abigail
+If I knew how to use a sword I would be exploring those caves by the mountain lake right now. It's probably nice and cool in there.;Abigail
+Ew, I hope I don't get a tan this summer.;Abigail
+I kind of wish I had a cat. Unfortunately, my Dad is allergic to pretty much everything.;Abigail
+I'm looking forward to fall... the cool mountain breeze, the swirling red petals, the smell of mushrooms... ;Abigail
+Do you ever get an urge to go exploring, <PLAYER>? pretend you just won a free vacation. Where would you go?;Abigail
+I do... but you probably wouldn't like the places I'm thinking about... Maybe we'll be neighbors someday. I think we both have similar, weird tastes.;Abigail
+Huh. I think that would get old fast.;Abigail
+Yea! That sounds wonderful.;Abigail
+I would love to live in an old farmhouse like you. The creak of a floorboard, the slow plumes of dust whispering through the rafters... There's something special about an old building, isn't there?;Abigail
+You're an interesting , <PLAYER>. I'm glad you moved here.;Abigail
+I'm actually enjoying myself this summer. It's funny how quickly one's outlook can change.;Abigail
+I used to think Pelican Town was kinda drab, but lately I'm starting to love some things about it.;Abigail
+Wow, you look nice today. I mean, that summer air is really good for the complexion, isn't it?... heh...;Abigail
+Do you ever get lonely on your farm?;Abigail
+Ah... Fall's here! I was getting so tired of summer.;Abigail
+I need to stretch my legs and get some fresh air today.;Abigail
+I try to help out in the shop on most days. Ever since Joja Mart shut down we've been doing great at the shop!;Abigail
+I can't wait to see some pumpkins this year. The spirit's eve festival happens at the end of the season. It's pretty low-key compared to what you'd see in Zuzu City, but for a small town like this it's a lot of fun.;Abigail
+Last night I dreamt that my left hand had turned into a gigantic <NOUN>. Does that mean anything? ;Abigail
+I can't believe you fixed up that old Community Center all by yourself. I'm really impressed. Do you have any scarecrows on your farm?;Abigail
+Hmm... what should I do tonight? I was thinking about dyeing my hair again... what do you think? Dye it black. Why not blonde? How about bubblegum pink? I like your hair just the way it is!;Abigail
+I was going to dye my hair but then I decided against it. Yes? What is it?;Abigail
+I guess that's a pretty safe choice, huh? I'll think about it.;Abigail
+Of course that's what you'd say.;Abigail
+You're always good for a laugh.;Abigail
+Are you growing pumpkins on your farm this year? Save one for me!;Abigail
+I may fight with my parents a lot, but I know they're just looking out for me. They're doing what they think is best. I dunno why I'm saying this to you. Don't tell anyone.;Abigail
+Hi. Your hair looks cool today... did you do something different?;Abigail
+You know, I could actually see myself becoming a farmer some day... But how would that ever happen?;Abigail
+I had an interesting dream last night, with you in it. It was a good dream.;Abigail
+I wish the season lasted a little longer. When the wind really picks up, I almost feel like a kid again.;Abigail
+I usually feel lonely this time of year, but not anymore. <PLAYER>, what are your plans for the future?;Abigail
+Well, fall is over... But I like winter, too!;Abigail
+It's just too cold to go outside much. But I do enjoy building a snowgoon.;Abigail
+Doesn't a steaming hot cup of cocoa sound good right now?;Abigail
+Hi. Is it boring to be a farmer during the winter?;Abigail
+I'm definitely going to enter the egg hunt tomorrow. Are you?;Abigail
+Well, tomorrow is the flower dance. Dancing is alright, but they make me wear this stupid white dress...;Abigail
+I'm looking forward to making a snow man tomorrow.;Abigail
+Will I see you tomorrow at the Spirit's Eve festival? Look for me in the haunted maze.;Abigail
+Everyone's excited for the Luau tomorrow. Maybe a little nervous, too, considering the Governor will be here.;Abigail
+You can't miss tomorrow's festival. It's one of the most beautiful wonders of the valley.;Abigail
+I'm going to eat so much cranberry sauce tomorrow.;Abigail
+Another year is almost over. It goes by so fast, doesn't it?;Abigail
+One thing I've learned living here... everyone stares at you if you look different.;Abigail
+It's the weekend! That means I can play twice as many video games. Bye.;Abigail
+Oh, hello. I made these two drawings. What do you think?;Abigail
+Remember those drawings I showed you a while ago?;Abigail
+Oh yeah? I guess that makes sense.;Abigail
+I'm glad. It's embarrassing to share your art with others. But I think you and I are on the same page.;Abigail
+Rude! Why would you hurt my feelings like that? You're so insensitive!;Abigail
+Hi. If you're ever bored, you can always hang out in my room. As long as I'm not changing, of course!;Abigail
+It's so cold, I wish we had a hot cup of cocoa to share.;Abigail
+I've been daydreaming a lot lately... Oh! You want to know what it's about? Well... ...It's a secret.;Abigail
+Another year is almost over... But this was a really good year, don't you think?;Abigail
+It must be nice not having crops to worry about this time of year. And that means you can hang out with me more often!;Abigail
+<PLAYER>, there's something I want to tell you... It's just... um... well... your boots! They, uh...look really clean... gulp That's all... sigh;Abigail
+I can't trust a person who doesn't like animals... Oh, don't worry! I know you're an animal lover!;Abigail
+Oh, hey. So you're the new guy, huh? Cool.;Alex
+You beat me to it! I was about to ask you... I'll see you out there.;Alex
+Aw yeah, that's the good stuff... Thanks for the birthday gift, <PLAYER>.;Alex
+What the...? Uh, I dunno what it's like where you grew up, but to me, this is trash.;Alex
+Whoa... this is great. Thanks!;Alex
+Aw, yeah... I've been looking for this one. It's supposed to be full of tips for bulking up. Rule number one... Eat a ton of eggs! Hah, no problem! Thanks!;Alex
+Uhhh... Reading? Yeah, um... I guess I can give it a try.;Alex
+Umm... I know I'm pretty big on eggs... But this isn't what I meant... Oh well... Down the hatch! gulp;Alex
+Can you eat this? I guess there's only one way to find out...;Alex
+Hehe... Keep giving me these every day, and I'll be able to retire at thirty!;Alex
+Now that's what I call an egg! Imagine the size of the omelet...;Alex
+Whoaa... It's so heavy I could use it for bicep curls. I'll be slurping on this for a week! Hah hah!;Alex
+Mmm... Duck egg... Rich and creamy. crack ... slurp . Ahhhh...;Alex
+Oh... um... I usually like eggs, but this one looks evil.;Alex
+crunch ... Mmm... Nuts. That hits the spot. Thanks, <PLAYER>!;Alex
+Hey, thanks. Eggs are great for bulking up. crack ... sluuurp ... Mmmm. Now that's what I call a snack.;Alex
+Have you been up by the lake recently? I was on a run late last night, and I noticed that the old bridge was repaired...;Alex
+Uh... I'm not used to opening up like that...;Alex
+Heard you caught some kind of rare fish? I'm impressed. Your arms must be stronger than they look! pinch ... not bad!;Alex
+Hey, I overheard someone saying that you got your first chicken the other day. Man... I wish Grandma had chickens. I'd eat a dozen raw eggs for breakfast each morning!;Alex
+Hey... uh... <PLAYER>? I... feel a lot more 'at peace' after talking to you on the beach last week. Thanks for listening to me.;Alex
+I saw you catch that salmon. If you happen to make a 'salmon dinner' out of it, let me have a bite!;Alex
+I saw you checking out the bath house the other day. Trying to get in shape? Heh. Uh... you didn't go poking around in my locker, did you?;Alex
+What's going on out there? Any action? I gotta stay here with my grandparents...;Alex
+I tried to go for a run this morning, but the plants have gone crazy! I couldn't get very far...;Alex
+Well, this is a season I'll never forget!;Alex
+You got Haley to marry you, huh? I always figured she was out of your league!;Alex
+Aw... I guess it can't last forever.;Alex
+I knew I shoulda brought a frisbee... Oh well.;Alex
+I wonder if anyone would be up for a volleyball tournament. Don't worry, I'll go easy!;Alex
+I wonder how far I could swim out, and still make it back? I'm guessing at least a mile. ... Don't worry, I'm not going to try it.;Alex
+Aren't you hot in that outfit? C'mon, man... It's the beach!;Alex
+With this island open, there's no excuse to be pasty anymore!;Alex
+Phew, it feels good to finally get these abs a little sunlight. Kind of a shame to have them hidden away all the time back home.;Alex
+When I go pro, I'll vacation on islands all around the world. But for now, this place will do just fine!;Alex
+I'm ordering a banana shake with two scoops of protein powder.;Alex
+Ah... The island life. I could get used to this!;Alex
+Okay, this should be fun... I hope I can remember the moves this year! It's been a while...;Alex
+Nah... I'm gonna ask someone else. Good luck.;Alex
+I thought we had something special... I guess I was wrong.;Alex
+You're not worth my time.;Alex
+I don't know if I can trust you anymore, but I'm willing to give you a chance.;Alex
+You're giving me wha...? huff ... Got it. Alright... well, you've made your choice. You're not gonna see me begging.;Alex
+Did you know I was an all-star quarterback in high school? It's true. See this little star on my jacket here? That proves it.;Alex
+The beach is a cool place to hang out and soak up some rays. You gotta spend some time in the sun or else you'll get all pale. sigh ... I wish there were more girls in this town, know what I mean?;Alex
+Looks like a good day to play catch, huh? I'd ask you to throw the ball around, but you don't really seem like the sports type.;Alex
+Go long! ...just kidding. Hahaha. I just wish I had someone to play catch with...;Alex
+Really? I guess I underestimated you. My arm's a little sore, but maybe next time.;Alex
+Yeah, some people just aren't made for sports.;Alex
+The air's starting to warm up... I'm feeling pumped.;Alex
+Hey, what's up farmer guy?;Alex
+My arms are really sore, but that's the sign of progress for a guy like me. I must've done a thousand push-ups yesterday.;Alex
+Hey. What, you wanna talk to me? I'm busy.;Alex
+Hey, <PLAYER>. That's right, I remember your name.;Alex
+Hey, <PLAYER>. How's your day going?;Alex
+Hey, <PLAYER>. I'm glad you stopped by. I'm not ashamed to say that I love my grandma! Now Grandpa, on the other hand... Just kidding.;Alex
+Nothing's better than an energetic day at the beach with your friends... right?;Alex
+You know, I used to want fame and fortune, but lately I've been starting to sing a different tune. In the end it's the humble little things that satisfy, don't you think? I still want to go pro, but it's not the most important thing in the world.;Alex
+Hi <PLAYER>. You look sporty today.;Alex
+I've been having a hard time staying focused lately.;Alex
+Of course you do. If you didn't I'd think you were crazy. Or some kind of nerd.;Alex
+Really? You must be some kind of weirdo. Sorry.;Alex
+No, I'm not! You're just trying to flatter me.;Alex
+Yeah, I guess you're right...;Alex
+You were supposed to disagree with me.;Alex
+Yes... I know. That's why I've been so determined to be successful with my sports stuff... Make it or not... As long as I try my hardest I think she'd be proud. Thanks.;Alex
+Wow... It's the anniversary of my Mom's death! Have a heart...;Alex
+...Thanks, <PLAYER>. You... you're different than the other guys. More sensitive.;Alex
+That doesn't make me feel any better, jerk.;Alex
+That's an interesting thought...;Alex
+They would never do that!;Alex
+...Maybe you're right... I dunno.;Alex
+Hey. Summer is definitely the best time of year. Right now is great because we've got the whole season to look forward to. Catch you later.;Alex
+Hey, farmer guy. You look a little burnt.;Alex
+If my hair wasn't so popular with the ladies, I swear I'd shave it all off in a second. sigh ... Life can be tough.;Alex
+Hey farm guy, I've got a question for you.;Alex
+Hey, when I turn pro you can say you were my first "fan". Not bad, huh? Hey, I've been practicing twice as hard so I can turn pro and prove you wrong. You just wait.;Alex
+Hey, now that's what I'm talking about! You're cooler than I thought.;Alex
+Yeah, well I'm so good I'm not going to have to practice all that much.;Alex
+That's insane. You're just jealous that I'm talented and popular and you're not.;Alex
+Hey, you must be getting pretty strong working on that farm all day.;Alex
+What do you want? I've got more important things to do right now.;Alex
+Mmmm... I smell a barbeque. Damn I could go for a burger.;Alex
+I got these new shoes yesterday 'cuz my old pair had a brown smudge. I just threw them into the garbage. I would've donated them but I don't like the idea of some weirdo wearing my shoes, ya know? ...What?;Alex
+Oh wow...your shoes are a little dirty... but that's fine, too! Different people have different tastes I guess.;Alex
+Hey, you look like you're getting into some good shape this summer! All that farm work is paying off, huh?;Alex
+Hey, nice tan.;Alex
+Sorry my room is so messy. It's just hard to always remember to clean. I'm trying to get better, though.;Alex
+You know, I actually wouldn't mind being a farmer... it seems a lot like playing sports, in a way. I like being outdoors and doing things with my hands.;Alex
+I've been trying to do more reading lately... I feel like I've been neglecting my brain for years because of my athletic obsession.;Alex
+Sure, it's getting colder... But it's still warm enough for sports! That's all I care about.;Alex
+My grandma told me I should spend more time studying. I was like, '...Grams, don't fret. I'm turning pro!' Studying is for nerds, anyway.;Alex
+Hey, I want to ask you something.;Alex
+I put too much gel in my hair today. It kind of hurts. Don't touch it! ...what?;Alex
+Yeah, I know. That's why I have it like this.;Alex
+Yeah right. You're just jealous that I look so good.;Alex
+What's up? Isn't your name, like, <FIRST_LETTER><NAME>, or something? Oh. It's <PLAYER>? Okay. Sorry.;Alex
+I'm going to do two hundred sit-ups tonight. I'm pretty dedicated, aren't I? Need anything else?;Alex
+I think you can find salmon in the river this time of year. At least that's what I heard.;Alex
+Hey, the gridball game is on today! Sundays are pretty awesome in the fall.;Alex
+What's up, <PLAYER>?;Alex
+I'm going to do a ton of sit-ups tonight. You have to work really hard if you want to achieve your maximum potential.;Alex
+My grandma told me I should spend more time studying. Maybe she's right...;Alex
+I finally took the plunge... I've decided to stop putting gel in my hair. I wouldn't want to be friends with someone who only liked me because I have gel in my hair.;Alex
+Winter is kind of lame. There's too much snow to do anything fun. I guess some people like it.;Alex
+You like the snow? Me... I'm just looking forward to spring.;Alex
+Hey, can you go find Haley and tell her I said 'hi'?;Alex
+Oh, hey. How's the married life coming along?;Alex
+People stay indoors a lot during the winter, and do boring stuff. Maybe if I read a book I'd be less bored... nah.;Alex
+When I turn pro, I'll be so rich that I'll get a vacation home in the tropics. Then I won't have to deal with these boring winters.;Alex
+You came all the way through the snow to see me? I guess that makes sense.;Alex
+There's some weird people living in this town. ...like that guy Sebastian. Why does he wear black all the time? I don't get it.;Alex
+Hey, the gridball game is on today! This might be my favorite day of the week.;Alex
+You came all the way through the snow to see me? That's nice of you.;Alex
+I guess it's interesting that some people are totally different than us, you know? I don't always understand why others act the way they do, but that's fine.;Alex
+Hey, <PLAYER>. Could you do me a favor? If you see Haley, tell her I'm busy. Thanks.;Alex
+I'll admit... I was a little jealous when you and Haley got married. But, hey... It's a chance for me to learn something new about myself.;Alex
+You know what? I'm going to start reading a book!;Alex
+I wish I could control the weather.;Alex
+If I ever make a lot of money, I'll make sure all my friends and family are taken care of. That means you, too.;Alex
+Er... hi. I'm the town blacksmith. If you ever need to upgrade your tools, I'm your guy.;Clint
+I don't usually get birthday gifts. This actually means a lot, <PLAYER>. Thank you.;Clint
+Is this a birthday gift? It's, uh... an interesting choice. Well I appreciate that you remembered me, anyway.;Clint
+You're giving this... to me? I don't think I deserve it, but.. thank you.;Clint
+Oh, interesting... These books aren't very easy to come by in the valley. Looks like there's a lot of good tips in here. Thanks, <PLAYER>.;Clint
+Ah, I can tell from the scratch marks on your pickaxe that you've reached a deeper layer of the mines. If it starts getting difficult to break stones, you'll want to upgrade your pickaxe. I can help.;Clint
+It's situations like this where a person's true character is put to the test. I'll keep Em... I mean, I'll keep everyone safe from the toxic rain!;Clint
+Well that was a dud. And here I thought it was finally my chance to be a hero. Oh well... Back to the grindstone.;Clint
+I feel bad admitting it, but sometimes I wish there was a real disaster, not just some dumb green rain. Something to shake everything up, you know? Not that I want people to get hurt... I just want to feel alive... To feel like I have a purpose...;Clint
+Interesting. There's tiny iridium flakes on your pickaxe... you've been going to the desert caves, huh? I've always wanted to smelt with iridium. It's not easy to come by.;Clint
+Did Emily say anything about our date the other day? Wait, don't tell me... I'm better off not knowing. Maybe I can still salvage my dignity.;Clint
+I... um... heard that you and Emily are together. Congratulations.;Clint
+I've been feeling strangely hopeful lately.;Clint
+I'm gonna put on my special red trunks. They're supposed to be really eye-catching.;Clint
+Well, that didn't turn out like I was imagining...;Clint
+Oh no... I forgot to clip my toenails.;Clint
+I thought working a furnace would have prepared me for this heat, but I'm sweating in places I didn't know I could. I hope no one noticed�;Clint
+Ow... my whole torso is burnt to a crisp.;Clint
+sigh ... We're all at the beach but I still feel so alone...;Clint
+Hmm... That's an interesting stone, like nothing I've ever seen before. I wouldn't know what to do with it, though. I deal in the mundane, not the magical...;Clint
+Somehow, the old town minecart system started working again. It's very convenient for getting to and from the mines. Don't get too cozy, though... if it can start up so suddenly, it could just as well shut down...;Clint
+Yep. I'm a blacksmith. My father was also a blacksmith. My grandfather was a blacksmith as well. I bet you can't guess what my great-grandfather was�;Clint
+And yes, my great-grandfather was also a blacksmith.;Clint
+Wow. That was rude. I was just being funny and you took it to a whole new level.;Clint
+Be glad you're a farmer, <PLAYER>. It's nicer to work outdoors than by a hot furnace all day. I'm only a blacksmith because my father pushed me into it.;Clint
+If you want me to upgrade your tools you'll have to give me the correct ore. And you'll have to pay me a fee, of course.;Clint
+Oh, hi <PLAYER>. Sorry If I look a little messy. I've been working at the furnace all morning. So, did you need anything, or do you just want to visit?;Clint
+Business has been slow lately. You should upgrade your tools. I could use the cash.;Clint
+I heard Emily has a friend who lives in the desert. Do you know anything about her? I wonder if she's single...;Clint
+The weather doesn't really matter to me. I typically stay near my shop year-round. Depressing, huh?;Clint
+I should save up and buy a computer... Maybe I can find a girlfriend on the 'internet'... What? Don't judge me! Not everyone is a cool, attractive, socially-adept farmer, you know! What's a guy like me supposed to do? Groan;Clint
+Sometimes I wonder how I ended up in this town.;Clint
+Damn it, I'm too old to do anything exciting anymore. All I have to look forward to is years and years of sweaty, hot furnace-work.;Clint
+How are your tools holding up? If they ever need repairs, just come by my shop. I guarantee these tools for life!;Clint
+So how are your tools holding up?;Clint
+I don't know if people realize this, or care, but I do take a lot of pride in my work. I mean... When I was a boy I didn't really dream of becoming a blacksmith. But that's what I am. And I want to be the best blacksmith I can be.;Clint
+Okay... I'll keep that in mind.;Clint
+That's the problem... I do act natural... but I never have any success.;Clint
+I hate to say it, but sometimes I'm glad my father forced me to become a blacksmith. At least I have a profession. If I didn't have work to do, I'd probably spend all day just sitting around feeling sorry for myself.;Clint
+Greetings! I'm a local scientist and father. Thanks for introducing yourself! I'm studying the local plants and animals from my home laboratory. Have you met my daughter? She's interested to meet you.;Demetrius
+Wow, this is a rare substance. I'll have to take it to the lab for... uh... further analysis.;Demetrius
+A present!... with optimal timing. Today is the anniversary of my birth. Thank you!;Demetrius
+A present!... with optimal timing. Today is the anniversary of my birth. Let's see what it is... ...Oh.;Demetrius
+Wow, an intact dinosaur egg! This is an extremely rare specimen. Now, if only it could be incubated...;Demetrius
+I already spotted a rare crab. This place is great.;Demetrius
+Don't mind me, <PLAYER>. I'm just collecting some samples. This may be a once-in-a-lifetime opportunity! Erm... If you see my wife, can you tell her I'm okay?;Demetrius
+The rain appears to be harmless, so I'm taking less precaution this year.;Demetrius
+I heard you visited Calico Desert recently. How fascinating! I'm sure you observed many unique life forms on your trip.;Demetrius
+Checking up on the melon research? Yes, I... uh... 'analyzed' it, and it certainly has an optimal macronutrient profile...;Demetrius
+Say that again!? You successfully incubated a dinosaur egg? This is unprecedented! Here, take my coffee, I'm all worked up now... I can't drink any more.;Demetrius
+Yes, Maru told me the news. I can't say I'm thrilled about it, but maybe you'll prove me wrong.;Demetrius
+sigh ... <PLAYER>, I've come to accept that it's time for my 'little girl' to become a woman. You have my blessing.;Demetrius
+Congratulations on the marriage. Robin's been crying since the wedding. 'Tears of joy', as she calls it! A most peculiar concept... Hmm, now that the basement is unoccupied, I might have to expand the laboratory...;Demetrius
+How's that cave experiment coming along? Has it produced anything of value yet? Ah! That means the conditions must be more suitable than I originally theorized... hmm..;Demetrius
+Don't forget to check that farm cave! There's bound to be something useful in there by now.;Demetrius
+I'll be analyzing this data for months.;Demetrius
+crunch ... whoops, there goes an endangered crab species... ... Oh, never mind. It was just the empty shell! I can sleep tonight.;Demetrius
+I usually don't partake... but I'm usually not on a tropical island, either. Hahaha.;Demetrius
+Can you believe this place? It's teeming with exotic life!;Demetrius
+It's a good specimen, but a very common species. You can keep it.;Demetrius
+Fruit and salt? What an odd combination...;Demetrius
+Robin won't even tell me who paid for Pam's new house! I pretend I'm mad at her but secretly I'm glad she's so loyal to her word�;Demetrius
+Maru helps me out in the lab sometimes... She's a good kid.;Demetrius
+It's hard to do good science without Maru's help... But her happiness is much more important than my experiment!;Demetrius
+This valley has a very vibrant and diverse ecosystem. That's one reason I was excited to move here.;Demetrius
+You're probably growing a lot of interesting plants on your farm, huh? Maybe I'll stop by your place some day and check it out.;Demetrius
+How's the farming business going? It's parsnip season, isn't it? I can imagine it being pretty peaceful, working outdoors with plants all day.;Demetrius
+Robin has a hot temper. It's better to stay on her good side. Don't tell her I said that.;Demetrius
+Let's see... If compounds in the rhizosphere contain sufficient levels of Carbon-13, then... Oh! Sorry. I was pondering some data and I didn't notice you there. Do you need anything?;Demetrius
+It's good to take a break from work every now and then. I guess that's kind of difficult when you live on a farm, though. Hey, at least in the winter you don't have to worry about crops.;Demetrius
+Robin has a hot temper. It's better to stay on her good side. ...now if I could only quantify her mood levels, perhaps I could model the fluctuations using a sinusoidal curve...;Demetrius
+Hi <PLAYER>! How are you doing today? I wish I could make your farming job easier somehow.;Demetrius
+Really? I figured a farmer would know the correct answer...;Demetrius
+See? <PLAYER> agrees with me.;Demetrius
+Well, we'd better titrate this solution before it crystallizes on us.;Demetrius
+Make sure to water your plants every day. Although, if you're using water retaining soil you don't need to worry about watering as often. It's pretty convenient.;Demetrius
+How's the farming business going? It's tomato season, isn't it? I can imagine it being pretty peaceful, working outdoors with plants all day.;Demetrius
+So I heard Maru had you look through that telescope out back. If you and Maru become friends, I'm sure she'll show you how to use that telescope out back. Pretty exciting, huh?;Demetrius
+I hope my daughter's keeping up on her math and science!;Demetrius
+Hmm... If strigolactone levels could be increased, would it have a proportional effect on mycorrhizal growth? Oh! Sorry. I was pondering some data and I didn't notice you there. Do you need anything?;Demetrius
+I became a scientist because I wanted to contribute to the progression of our species. Of course, none of this would be possible without farmers such as yourself.;Demetrius
+During this time of year I divert my attentions to fungi.;Demetrius
+We know a lot about nature thanks to science. But there's always a lot more to learn.;Demetrius
+How's the farming business going? It's corn season, isn't it? I can imagine it being pretty peaceful, working outdoors with plants all day.;Demetrius
+Maru sure spends a lot of time working on those gadgets of hers. I'm glad she's so dedicated.;Demetrius
+I'm still flabbergasted about Maru's secret robot... was it all a dream?;Demetrius
+Let's see... perhaps the nitrogen fixation level is affected by nematode secretions... Oh! Sorry. I was pondering some data and I didn't notice you there. Do you need anything?;Demetrius
+Hi <PLAYER>. I'm always glad when you come for a visit. It's nice to have some company after all this lonely lab-work.;Demetrius
+I'm taking a break from my research right now.;Demetrius
+Talk to my wife if you want to increase the size of your house.;Demetrius
+Do you have a kitchen in your house yet? Cooking is kind of like doing lab work.;Demetrius
+Do you have any animals on your farm?;Demetrius
+Let's see... if seasonal nitrogen levels could be modeled by a piecewise function... Oh! Sorry. I was analyzing data and I didn't notice you there. Do you need anything?;Demetrius
+Feel free to harvest as much lumber and stone from this area as you please.;Demetrius
+Thanks for stopping by. I need to think about something other than legumes for a while. So, what have you been doing lately? That sounds interesting.;Demetrius
+Hey, you're pretty good at mining, for a human.;Dwarf
+Oh! I've read about this tradition... it's a 'birthday gift', isn't it? Hmm, very interesting...;Dwarf
+Oh! I've read about this tradition... it's a 'birthday gift', isn't it? So, is it part of the tradition to give out garbage like this?;Dwarf
+I can't see the color green very well. I guess it's part of living underground for so long.;Dwarf
+sniff ... is this milk? sip ... hmm... tastes good.;Dwarf
+I got all this stuff from the surface. I just take it from your people during the night. Hmm? What is this concept you call "private property"?;Dwarf
+Ah, 'milk'... slurp ... How interesting. Remind me, again... where exactly does this juice come from? ...!!! sploosh;Dwarf
+Ah... cave carrot stew, cave carrot roast, cave carrot soup and cave carrots au gratin. ...And a ramekin of cave carrot brulee to sweeten the night. What?;Dwarf
+If you enter the volcano on Ginger Island, you might encounter others like me. Don't be alarmed! We're peaceful with humans.;Dwarf
+You know, you're pretty good at mining, for a human.;Dwarf
+Oh, you caught a Stonefish? I know them very well. We sometimes keep them as pets.;Dwarf
+Oh, you caught an Ice Pip? They're energetic little fish, despite the frigid environment they live in. My people avoid the frozen areas, though... too cold!;Dwarf
+Oh, you caught a Lava Eel? Very good! Lava Eel fishing is a big part of our culture. They're considered a delicacy!;Dwarf
+You found the legendary Dwarvish Forge? Incredible... I've never seen it, myself. The machinery is so old, none of my people understand how it works. But it's definitely Dwarvish technology.;Dwarf
+Hey, wanna buy something?;Dwarf
+Long ago, my people knew the secrets of advanced technology. The archaeological evidence proves that. But I wonder where it came from? And where did it all go? I guess some questions will never be answered...;Dwarf
+I feel most comfortable underground. The deeper underground the better.;Dwarf
+You've never seen a dwarf before, have you?;Dwarf
+The mines can be a treacherous place. Be careful, friend.;Dwarf
+It's dangerous in the mines. Why don't you buy a few things from my shop to ensure your safety?;Dwarf
+One day you'll have to tell me about this beverage you call "milk".;Dwarf
+You're not a spy sent by the Shadow People, are you? I'm keeping my eye on you.;Dwarf
+Most humans fear the unknown... that's why I've hidden my shop here. Your mining skills prove you're trustworthy.;Dwarf
+Ah, the new farmer we've all been expecting... and whose arrival has sparked many a conversation! I live in the little cabin by the beach. It's a pleasure to meet you.;Elliott
+I'm honored that you would remember my birthday! Thank you!;Elliott
+Another year gone by, another gray hair... I suppose this gift can commemorate my decline.;Elliott
+Ah, what a gift! The aroma alone is inspiring...;Elliott
+Of course... it would be an honor! I look forward to it.;Elliott
+Ah, you've been doing some beach combing... A fine hobby! Thank you.;Elliott
+Ah, a book... Yes, perhaps the prose within these pages will offer a new insight. Thank you.;Elliott
+Ah, a bottle of fine ink. A writer can never have too much... and it's quite expensive! Thank you!;Elliott
+This will make a beautiful quill! I feel inspired already...;Elliott
+It's rare to see so many people at the beach. All for the love of squid!;Elliott
+Ah... please excuse the mess in my cabin. I wasn't expecting anyone to enter!;Elliott
+Agh, it's still wriggling! Get that abomination away from me!;Elliott
+I only wish I had given you a bouquet first!;Elliott
+You've plucked Pelican Town's finest flower... Now, you must care for it with all your heart.;Elliott
+Well, if it isn't the town's number one ice fisher! That was truly an impressive performance. I never stand a chance, but I always participate just to be sporting.;Elliott
+Did you know? A duck's feather makes for an excellent quill.;Elliott
+You've been very helpful... Now, I must think...;Elliott
+I had a good time at the saloon the other day... Though my liver is not quite so enthusiastic...;Elliott
+That was quite the ride!;Elliott
+I'll see you back in the valley.;Elliott
+Hopefully a crab hasn't snuck onto my clothes again. Imagine crawling out of a pocket and finding yourself on a beach hundreds of miles from home! Hmmm... Perhaps this could make for an interesting novel...;Elliott
+I'm just taking a quick break to slather some coconut oil on my body.;Elliott
+Gus, I'll order a round for everyone! hic ... Oh... I don't have the G. Heh heh... Never mind!;Elliott
+I've heard rumors of pirates coming to shore near here at night. They say it can get pretty wild!;Elliott
+I must admit, this beach blows my own humble strand out of the water.;Elliott
+Oh dear! Some seaweed has tangled itself in my hair. That's the trouble with relaxing on the beach.;Elliott
+Yes... could I refuse that soft, kind face? The touch of spring-time's sweet embrace?;Elliott
+Excuse me... not today.;Elliott
+... Why do you torment me? Can't you see you've shattered all my hopes and dreams?;Elliott
+Go away. I can't bear to see you.;Elliott
+You've hurt me beyond measure... but I'm willing to leave the past behind us.;Elliott
+Ah... the wilted bouquet. A sight I'd hoped to never see... I suppose there are other horizons to set my eyes upon...;Elliott
+I'm kind of new to this town myself, but I really feel at home. I moved here only a year before you.;Elliott
+I hope you've come to think of this place as 'home'.;Elliott
+A great idea can pass through your head when you least expect it... but if your mind is too busy you might miss it. Well, I really must get back to my work.;Elliott
+The sweet friction of pen and paper is the music of my soul. That's why I chose this beach as my home, so that I could have peace and quiet to do my work.;Elliott
+The forest is a wonderful place. Have you been there?;Elliott
+Sorry, I'm feeling morose today. I wish there was more certainty about the future. I don't want to grow old as a lonely hermit on this beach...;Elliott
+I can't seem to find the inspiration to begin writing my novel...;Elliott
+I've been feeling hopeful lately. Perhaps the weather is changing.;Elliott
+Sometimes I wonder if I might just have an inflated self-image and no real skills... No, no... I'm not fishing for compliments. Though they are appreciated...;Elliott
+It's a little lonely out here on the beach... so I apologize if I was ever a little too forward with you when we first met. I was just eager to have a friend. It feels good to have a close friend like you.;Elliott
+Hello, <PLAYER>. Are you well?;Elliott
+The fresh air of this valley is good for body and mind. A quick stroll outdoors always invigorates me.;Elliott
+Ah... do not get too close to my hair with a torch...;Elliott
+I'll admit... it takes me several hours each morning to make my hair look this good.;Elliott
+You probably wouldn't like it inside my cabin. It's dark and full of spiders.;Elliott
+Please excuse the sorry state of my cabin.;Elliott
+Some people are shy. Keep showing interest in them and they'll get comfortable around you. Everyone likes to have friends, even that grumpy blacksmith.;Elliott
+Hello. I hope your new farming life is panning out as you'd hoped?;Elliott
+Oh, <PLAYER>! I was hoping you'd show up. It's always a pleasure to see you.;Elliott
+Mystery, huh? It's definitely an exciting genre. I'll remember that.;Elliott
+Ah, one of the classic genres. I'll remember that.;Elliott
+I would've never guessed! I suppose even those of the 'earthiest' profession sometimes have their heads in the stars. I'll remember that.;Elliott
+... I was worried you might not feel this way about another man.;Elliott
+That's a great idea! Here's to us!;Elliott
+Ah, to a harmonious future for the community... what a virtuous idea. Here's to Pelican Town!;Elliott
+A gentle little sunbeam woke me up this morning. I've never felt so refreshed. I'm sorry if that isn't very interesting to you.;Elliott
+I know that I am kind of an 'oddball'. I hope you don't mind.;Elliott
+<PLAYER>, I know you have faith in my abilities as a writer. That means a lot to me.;Elliott
+I've encountered some beautiful shells in front of my house. I would imagine the rarer varieties to be quite valuable.;Elliott
+I write in hopes of connecting with others through time and space. sigh ...times are changing, though. People don't read anymore.;Elliott
+It must be satisfying to follow your crop from seed to harvest. It's as if your essence is infused into the fruit.;Elliott
+Oh dear! My shoes are filled with sand. That's the trouble with living on the beach.;Elliott
+<PLAYER>? You look puzzled. I guess the hot summer air can make us a little dizzy.;Elliott
+The sun is angry today... My skin is a bit too delicate, I'm afraid.;Elliott
+You came all the way here just to talk to me? How kind.;Elliott
+Visit my cabin whenever you like. I could use the company.;Elliott
+Hello <PLAYER>. Make sure and take breaks from your work now and then.;Elliott
+Come visit me if you need a break from your labors. I'm usually at my house.;Elliott
+It's been said that a pirate's ship, full of plundered gold, shipwrecked here a long time ago.;Elliott
+Let's not get in a tizzy, now... Gus! I propose a round of drinks to settle the nerves!;Elliott
+Ah, yet another unique day in the valley. Stimulating, to say the least!;Elliott
+<PLAYER>, I was just thinking about you. Maybe you sensed it. So! Tell me about your day.;Elliott
+I wouldn't mind trying my hand at farming. The quiet atmosphere of a farm might be a good source of literary inspiration, don't you think?;Elliott
+Oh dear! A tiny crab appears to have made his home in my shirt pocket. That's the trouble with living on the beach.;Elliott
+My legs are stiff from sitting at my writing desk all night. Sometimes I envy you, <PLAYER>.;Elliott
+My legs are stiff from sitting at my writing desk all morning.;Elliott
+<PLAYER>, when you're around... I feel unusually creative. I know... strange, but it's true.;Elliott
+It gets windy this time of year. Marvelous!;Elliott
+I have to remember to water my plants today. And not with sea water this time!;Elliott
+You've come to visit me in the cold? I could do with a respite from my lonely labors.;Elliott
+<PLAYER>, you've trudged through the snow to visit me? I'm honored!;Elliott
+People have scraped a living off the sea for thousands of years. I just go to the grocery store.;Elliott
+<PLAYER>, I had a feeling that you would show up. Perhaps we're connected by an other-worldly thread.;Elliott
+A day can never be boring when you follow the whims of your imagination. Sorry, am I babbling on about nonsense?;Elliott
+Breathe deeply. Do you notice it? That's the smell of the sea. How does it make you feel?;Elliott
+Breathe deeply. Do you notice it? That's the smell of the sea. Whenever I smell the sea, it reminds me of my youth. The ocean really impressed me as a child.;Elliott
+I have to brush my hair daily, or else it'll clump up into messy knots. It's a lot of work. I'm surprised I haven't just shaved it off in a fit of passion. I suppose I am too vain.;Elliott
+I've been doing little indoor exercises, since it's often too cold to go out. Sorry if it's a little humid in the cabin.;Elliott
+I really should scrub my floorboards today. I think an algae is starting to form. Do you make time for cleaning?;Elliott
+Ooh!... I can read it on your face. You're going to love it here in Pelican Town. If you're ever looking for something to do in the evening, stop by the saloon. That's where I work!;Emily
+Oh, you remembered my birthday! Thanks, <PLAYER>! That's really nice of you.;Emily
+Oh, you remembered my birthday! Thanks, <PLAYER>! That's really ni-- ... sniff ... sniff ... what is that smell? ... What is this...?;Emily
+Yes, I'd love to dance with you, <PLAYER>! Now I'm excited... This is going to be a lot of fun!;Emily
+Wow, thank you! I've heard that if you drink this before bed you'll have vivid dreams.;Emily
+Oh, good, it's still warm... Poor thing, where's it's mother? Whisper ...I'll be your mother, don't worry... Poor baby. I'm glad you gave this to me, <PLAYER>. I'll take very good care of it.;Emily
+Oh, wow... Thank you! This color perfectly captures my mood today. The warm, soothing energy of this stone helps us re-center our thoughts... It's actually just what I needed on a day like this. Thanks!;Emily
+That's funny, I had a dream that you would give me a piece of jade today! Ooh, it's jade! This is supposed to attract wealth... And who couldn't do with a few extra coins in the pocket?;Emily
+Oh, it's beautiful, <PLAYER>! I hope you didn't put yourself in too much danger to get this. Ah, an amethyst! This will protect me on those late nights working at the Saloon.;Emily
+Wow, look at that color! It matches perfectly with my outfit. Oh, this is perfect... I've been feeling a little sluggish lately. Rubies are great for energizing the spirit!;Emily
+It's so smooth! When I hold it... I can feel the energy of all the mermaids in the sea. Thank you! Thank you! I'm gonna put this under my pillow, and hopefully I'll dream about the ocean.;Emily
+Oh!... Heheh... I know exactly why you gave me this stone today. Very interesting... Wow, it's beautiful! This color makes me think of all the spiritual power within the forest... Can you feel that wood energy?;Emily
+There must be a meaning behind this... Nature is trying to tell us something. What, though? I can't tell... It's hard to read. I feel a strong presence, though...;Emily
+Can you feel it? The forest is singing! The trees are dancing with joy!;Emily
+That was a beautiful wedding... I'm really happy for you and Haley!;Emily
+Haley is too proud to be heartbroken over the divorce... Or at least to show it.;Emily
+Hello <PLAYER>. I had a feeling I'd be seeing you today.;Emily
+Nothing is more sad than an injured animal...;Emily
+Now that you have a sheep, you'll be able to make cloth! That'll open up a whole new world of tailoring possibilities. Just wait and see.;Emily
+Have you ever tried cooking without a recipe? It's kind of liberating. And who knows... you might discover something amazing!;Emily
+My outing with Clint was pretty fun! Although, he spilled a root beer all over himself within ten minutes of getting there, and then barely said a word the rest of the night.;Emily
+Hey, I saw you at Sam's show. It was fun! The city smelled pretty bad, though.;Emily
+I won't forget our camping trip any time soon...;Emily
+We're going to have a lot of fun together, I just know it.;Emily
+Well, now that you're married, I assume Shane won't be coming to the Saloon as much! That's probably a good thing, but it's always a little sad to lose a regular...;Emily
+Oh, I like this place! It's full of parrots!;Emily
+Ah... my lungs feel perfect.;Emily
+Can you feel it? The ocean has a profound healing energy.;Emily
+It's nice to be on this side of the bar for once!;Emily
+If you open your heart to the sun, you can feel a loving energy inside of you.;Emily
+Of course, my love. I've been looking forward to this!;Emily
+I'm very sorry... but I've already agreed to dance with someone else.;Emily
+<PLAYER>... You broke my heart... I just can't talk to you anymore. Please, let me try to heal...;Emily
+B...But I thought we had a special bond... sniff;Emily
+Oh, that's beautiful... I have a friend who would love that! But she lives far from here...;Emily
+<PLAYER>... I'm ready to give you a second chance. I still have faith that there's a good person inside of you.;Emily
+Have I told you Haley and I are sisters? Strange, isn't it?;Emily
+I have a secret hobby, but I won't say any more. Maybe I'll show you some day.;Emily
+This world is full of spirits and magic... Some don't believe me, but I know it's true. I can see it in your eyes... you believe in the other world, like me.;Emily
+I wish Haley would get a job or at least contribute to cooking and cleaning. I think she's hoping to marry someone rich.;Emily
+I have to say, now that Haley's gone I've had a much easier time keeping things in order.;Emily
+There's good people in this town who just can't seem to find happiness. It makes me sad. Working in the saloon, I hear first-hand about everyone's problems.;Emily
+I work part-time at Gus' saloon. It pays the bills.;Emily
+I'm just working at Gus' to make ends meet... but my real passion is tailoring. I made these clothes from scratch, see?;Emily
+Oh, hi... It's nice of you to visit me like this.;Emily
+You know, <PLAYER>... I think the future looks bright for both of us.;Emily
+I've heard rumors of rare and powerful magic rings, forged long ago by forgotten civilizations. I'm not sure if it's true or just a fairy tale.;Emily
+You catch fish? I just hope you treat the poor things humanely. Every critter deserves our respect, even the slimy ones.;Emily
+Have faith... every day you're taking one more step on the path of your destiny.;Emily
+Hey, <PLAYER>. I always knew we would become good friends. It was written in the stars!;Emily
+I hope you're farming in a sustainable way. The valley's ecosystem is fragile! How do I know? I have a strong gut-feeling!;Emily
+Sometimes the flowers speak to me... each one has a different story to tell!;Emily
+Ah, spring. The season of pastels. I actually prefer jewel tones, myself. Oh, excuse me! I was mumbling about fashion again, wasn't I?;Emily
+I like making my own clothes, but it's not easy to get cloth. And it's such a long trip to the city.;Emily
+Do you ever think about the texture of fabric, <PLAYER>? Each texture has a distinct personality... feel my dress... see?;Emily
+Hi <PLAYER>! Do you raise sheep in your barn? You can turn their wool into beautiful bolts of cloth! You'll need a loom, though. You can make them yourself if you know the crafting recipe.;Emily
+This house was left in our care by my parents. They've been traveling the world for the last two years. We have no idea when they'll be back. I enjoy living here, though. It's a beautiful area and the town is nice.;Emily
+Um... Okay. Sorry you didn't like it. I guess those three years of practice were a complete waste...;Emily
+Did you really think so? That's exactly the effect I was shooting for! I'm so glad you liked it.;Emily
+Haley is a picky eater. It's so hard to cook for her.;Emily
+Eating healthy is very important to me. I want to live a long and active life.;Emily
+Haley and I get along much better now that she's moved out. I look forward to her visits!;Emily
+I feel bad for Linus. People judge him, but he's just living a different lifestyle. I wish everyone could learn to be accepting of others.;Emily
+Hi, <PLAYER>! We've become decent friends, haven't we? That's fine with me. You're a cool guy.;Emily
+Hi, <PLAYER>! We've become decent friends, haven't we? This may sound weird, but I've always been certain this would happen.;Emily
+It doesn't rain much in summer, does it? That just means more watering work for you, huh? Or do you enjoy the sunny weather?;Emily
+I consider the bees and butterflies to be my special friends!;Emily
+If I wasn't so busy I'd come help you on the farm! But Gus would be upset if I had another job...;Emily
+You know, I'm sure Gus would understand if I wanted to leave the Saloon someday. There's so much out there, I can't stay a barmaid forever.;Emily
+Don't overwork yourself or you might end up in Harvey's clinic!;Emily
+It's around this time of year that my good friends visit the region... I mean the parrots!;Emily
+Well, fall is here. Time to plant some new crops, huh?;Emily
+I'm visiting Sandy today. We're best friends! I'm really happy that the bus is repaired. Sigh ... If only that old bus were fixed, I would've visited my friend Sandy today...;Emily
+It's good weather for a picnic, isn't it?;Emily
+If I didn't have to work I'd suggest we go on a nice picnic!;Emily
+Hi <PLAYER>! I bet your farm did great this past summer, huh? You're an excellent farmer.;Emily
+My friend runs a shop in the desert. If you ever go there stop in and say hi for me. I think she sells some rare products.;Emily
+Are you friends with Clint? He's a nice guy, if you get to know him. He gets lonely working in his shop all day. But he's so shy he has trouble making new friends.;Emily
+I think Clint's mad at me. He never looks at me anymore... I always thought we were friends.;Emily
+Another beautiful fall day. It's a pity it'll soon be over.;Emily
+Smell that clean air? It's lovely, isn't it? I just fear that the Joja Co.'s of the world will take this away from us.;Emily
+Keep your mind open and some interesting things might take root. Anyway, how is your day going?;Emily
+I don't have any huge plans for life. I prefer not to stress about the future.;Emily
+Nothing like a sunny Sunday to lift the spirits!;Emily
+There's so many invisible barriers holding us back. Why can't people learn to let go and explore their true potential? Me? Well, you know what they say... 'You can't cage a wild Junimo'.;Emily
+You're such a nice person, <PLAYER>! Thanks for being my friend.;Emily
+Be careful an icicle doesn't fall on your head!;Emily
+Be careful an icicle doesn't fall on your head! I wouldn't want you to get hurt!;Emily
+Oh, hello <PLAYER>! You must be getting very good at farming by now, huh?;Emily
+I bet you're an excellent farmer by now!;Emily
+I think my hair is frozen solid...;Emily
+Haley and I aren't made for such a big house. It's a hassle to keep it clean. Ah well. I really shouldn't be complaining. Your farmhouse seems nice and cozy.;Emily
+It must be so cold for the poor fish. I wonder if any fish hibernate? Maybe I should ask Demetrius.;Emily
+If I wasn't so busy I'd help you dig around in that cave by the lake. I bet there are some powerful crystals inside there.;Emily
+Do you know how to cook 'Bean Hotpot'? I like that dish quite a bit. It's hard to get fresh ingredients in the winter, huh?;Emily
+My sister and I might fight sometimes, but I would really miss her if she left. I think she'll become a very nice young lady some day... Don't tell her I said that. So, is there anything you want to talk about?;Emily
+Say hi to my sister for me! And remind her to visit me tomorrow.;Emily
+Why, hello and welcome to our little community, dear! You can call me 'Granny' if you like.;Evelyn
+Oh, you're giving this to me? Thank you so much, dear. This is a really special gift!;Evelyn
+Thank you, dear. I've been looking everywhere for my reading glasses... My, my! They've become filthy...;Evelyn
+George, dear... Is this the end? It's been a good life, but I'd like to see Alex find his way before departing...;Evelyn
+Oh, my dear... I'm so glad everything ended up okay. And my flowers look healthier than ever!;Evelyn
+The flowers seem to love this rain...;Evelyn
+Thank you, dear. This brings back such sweet memories...;Evelyn
+Oh, I'm sure George would be interested in that!;Evelyn
+Home-grown tulips make for a lovely springtime gift.;Evelyn
+It's lovely to have the old community center back. We have so many memories from the old days... It's a special place.;Evelyn
+I'm very happy for you, dear. Seeing young people in love is heartwarming for old grannies like me.;Evelyn
+Next time you marry, make sure it's the right one...;Evelyn
+That's a very pretty rose, dear. Have you shown it to Emily? It makes me think of her.;Evelyn
+Oh, sweetie... that was so nice of you to put Pam up in a beautiful new house! It warms my heart when people take care of each other like that...;Evelyn
+Call me a silly old fool... but I think I have an idea who paid for Pam's new house... wink;Evelyn
+Oh my! That's so kind of you, dear. Why yes... I'd love to go to the movies! You go on ahead, and I'll meet you at the theater.;Evelyn
+Don't mind my husband, George. He isn't very friendly to strangers. If you get to know him better he'll warm up to you. I'm sure you two could become good friends one day! Since you're so interested in my husband I'll let you in on a little secret: he really likes leeks! You can find them in the mountains this time of year.;Evelyn
+I saw the most beautiful family of butterflies sunning themselves in the town garden yesterday. The good Mayor put me in charge of the town's public gardens. I've been working on the town flower beds for weeks, so don't step on them!;Evelyn
+I'll be baking cookies today. I like to make flower shapes in the spring. Maybe I'll tell you my recipe one of these days.;Evelyn
+George spends the whole day in front of that darned television set. I wish he'd go outside more. Some fresh air would do him good.;Evelyn
+I used to love looking at the clouds, but I have trouble seeing them these days. My eyes just don't work as well as they used to.;Evelyn
+It's nice to be so close to the ocean. The sound of the sea makes going to sleep a lot easier. As kids we hunted for seashells after the tide went out. Those were the days...;Evelyn
+It's nice of you to take the time to visit George and I. Without young people around, we'd age much faster.;Evelyn
+What was that, dear? I could've sworn you said 'hockey puck'. sigh My ears aren't what they used to be.;Evelyn
+I feel more energetic in this wonderful weather.;Evelyn
+How's my grandson doing? I imagine he's quite happy living on the farm. He deserves to be happy... he's a good boy.;Evelyn
+George is always a little less grumpy in the summer. The warm weather softens his bones. You might not be able to tell, but I can. I've known him for sixty years!;Evelyn
+The good Mayor put me in charge of the town's public gardens. I've been working on the town flower beds for months, so don't step on them!;Evelyn
+I'm going to bake cookies today. They're coconut wonders shaped like little palm trees! Heehee. Maybe I'll tell you my recipe one of these days.;Evelyn
+Be careful out there, dear. You could get sunburnt!;Evelyn
+Sweet peas are in bloom right now. They've got such a wonderful fragrance, haven't they?;Evelyn
+It seems like we always have more visitors when I start baking cookies. My mother taught me the recipe.;Evelyn
+With every year I appreciate the seasons more and more. For young people it's easy to let them slip by.;Evelyn
+I've been putting raisins in George's oatmeal... but he hasn't said a word. I wonder if he even notices these things?;Evelyn
+The good Mayor put me in charge of the town's public gardens. The fall flowers are delicate, so don't go stomping on them! I know how you kids like to do.;Evelyn
+Today I'm baking cookies. They're shaped like little pumpkins. Maybe I'll tell you my recipe one of these days.;Evelyn
+Are you friends with Alex? Sometime he worries me sick.;Evelyn
+Oh, hello my dear. I think of you as my own now.;Evelyn
+My eyesight may be poor, but my nose hasn't failed me yet! I can smell fallen leaves and mushrooms. It reminds me of the forests I used to play in as a girl.;Evelyn
+Be careful you don't slip on the ice, dear.;Evelyn
+Aren't you cold? You should wear a thicker jacket, my dear! This weather isn't good for my joints...;Evelyn
+I usually tend to the gardens, but they're covered in snow. I still like to check up on them now and then.;Evelyn
+Today I'm baking cookies. They're shaped like little winter stars. Maybe I'll tell you my recipe one of these days.;Evelyn
+We've lived here for over 30 years. We've seen a lot of people come and go. Your grandfather, for one. Bless him... sigh;Evelyn
+Oh... my dear. You have your grandfather's eyes... Yes... yes... I knew him very well.;Evelyn
+You should make some hot soup to stay warm. And make sure you wear a jacket!;Evelyn
+<PLAYER>, have I told you that George and I have lived in our house for 30 years? I hope you stay here for a while.;Evelyn
+Hmmph... It's irritating to have to meet all these new people, huh? Now buzz off... Hmmph.;George
+Hmm... sip ... ahhh... Tastes just like Mama's tuna casserole! Delicious!;George
+What's this? Hmmph... so you push this button first? What's this button do? Hmmph... grumble Hehe! Well, how about that! It's a new channel! Here, take this. It's a stone I found when I was a coal miner back in the day. Funny thing, it's glowed blue and stayed cold for over forty years. Like magic!;George
+There, there, honey... Don't you worry. Even if it is poisonous, I breathed in more noxious gas in one minute back in the ol' coal mines... And I'm still kickin'!;George
+Looks like the young folks will have to do some extra weeding after this one...Heheh...;George
+Thanks for replacing that beat up old trailer... the new house is much nicer to look at. You know, I'll bet it increases the value of our property too!;George
+Welp... that new house blocks my view of the river a little... but it sure looks nicer than that old tin can Pam was livin' in.;George
+You wanna see a movie with an old geezer like me? ...Well... okay. I'll meet you there.;George
+I dunno what you did, but those raccoons haven't been back since last week!;George
+I may seem out of touch, but when you've been around for as long as I have, you learn a thing or two about the ways of life. It's called 'wisdom'. Young people don't understand the big picture. How could they?;George
+Hmmph... took the easy way out, huh? People these days treat marriage like a television set... If you get bored, just change the channel! Back in my day... grumble;George
+So... the old Community Center is gone, huh? Whatever happened to small-town values? It's a shame how times have changed...;George
+I'm not much of a talker. Especially not to strangers, if you don't mind me saying.;George
+Hmmph... Looks like another gloomy day.;George
+Alex is my grandson. Have I told you that already? Huh? Speak up.;George
+How can two men get married? It's unnatural... Hmmph. I guess I'm just "old fashioned"...;George
+I'll admit, I thought it was... strange... for two men to be together. But you're such a nice young man, and I know you two are in love... I've changed my mind.;George
+I can't talk right now, young man.;George
+So you're a farmer, huh? At least it's honest work.;George
+The weekend is no different than any other time, for me. That's how it is when you're retired.;George
+Aren't you cold? They don't make sweaters like they used to. Sigh;George
+I'm surprised you want to be friends with an old-timer like me.;George
+Hello, <PLAYER>. Wanna watch the TV with me?;George
+My grand-pap was a farmer. It's a respectable profession. If I wasn't so darned old I'd come to your farm and show you a thing or two...;George
+I wish my legs worked so I could play catch with Alex. That'll never happen.;George
+I've been thinking...Maybe I've been a little bit of a grouch. I used to think no one cared about us old folks.;George
+You know, I really have been a grumpy old man to everyone. But now I feel like I'm softening up like an old ripe fruit. Hehe! Take care, son.;George
+<PLAYER>! Let me turn off this darned television and we can have a little chat. So what are the young people doing these days?;George
+You know, I shouldn't get too depressed about my legs. I should just be happy that I'm alive.;George
+When you get as old as me it's a lot harder to adjust to new things. One day you'll know what I mean. Go on, enjoy yourself while you're still young.;George
+Ohh... It's so cold. When you're young the cold doesn't bite as hard.;George
+You don't like it, huh? Bah! Kids these days... no appreciation for anything. That was expensive, and I'm not made of money, you know!;George
+Well hello there! I'm the chef and owner of the Stardrop Saloon. Stop in if you need any refreshments. I've always got hot coffee and cold beer at the ready.;Gus
+Oh wow, thank you! I'm tempted to drink all of it, but I'm going to save some for the saloon. I bet I could make an interesting cocktail with this!;Gus
+sniff ... Ah. Exquisite cuisine. You know what I like, <PLAYER>.;Gus
+sniff sniff ... that sharp smell! slurp slurp ... that zesty taste! That gritty texture! It can only be gourmet tomato salt! I'll trade you for it.;Gus
+I saw the weather report last night. It really gave me a scare! Are there monsters out there? What's happening?;Gus
+I'm glad that's over. Running the saloon is enough to worry about, I don't need exotic weather in my life!;Gus
+sniff ... These rains have a particular smell. Have you noticed? Almost like freshly minced herbs...;Gus
+Lewis tells me your tomatoes have ripened. That's great news! ...Just imagine what the culinary world would be like without tomatoes...;Gus
+It's important to me that people can eat healthy, delicious meals. Unfortunately, in today's world, that's not so easy for everyone.;Gus
+Hey <PLAYER>, just want to say that I'm happy for you and <SPOUSE>.;Gus
+Emily has a real pep in her step lately... I wonder why? wink;Gus
+Hey, I'm sorry to hear about the break-up... If you need company, you're always welcome at the Saloon.;Gus
+Another failed marriage? Well, it happens. Guess I'll be seeing you at the Saloon more often...;Gus
+Hey, you really pulled through with those eggs. The omelet turned out great, thanks to you!;Gus
+Hey, <PLAYER>! Glad you're here. I'm gonna be serving up some good stuff from behind the bar.;Gus
+Meet me at the bar if you need a beverage.;Gus
+Hey, need a drink? I've got some special beverages that showcase local ingredients from this island.;Gus
+These might be the best crab cakes I've ever made...;Gus
+Yeah, I know a lot about the people living here. That's one of the benefits of being a bartender. Sometimes I hear too much...;Gus
+Hey. I heard what you did for Pam. That was really really nice of you! Sadly, I don't think it will get her to cut back on the 'beverages'...;Gus
+I saw Pam earlier today... she was grinning from ear to ear! She said someone bought her a new house! Whoever it was, they truly deserve the best!;Gus
+Welcome to the Stardrop Saloon! Can I get you anything?;Gus
+Hi there, <PLAYER>. Make yourself at home.;Gus
+Hi, Pierre! Now, what do I need� Ah, that looks good! Makin' my special sauce tonight! Pierre! What's fresh? Pierre! Waddya got for me?;Gus
+Hi there <PLAYER>. I'm glad to see you! You're always welcome here.;Gus
+Hi! Please relax and enjoy yourself.;Gus
+Hey, you look like you could use a beverage.;Gus
+Business has been really good tonight. I'm pleased.;Gus
+I sell different dishes each week, so make sure and check in every now and then! You might taste something spectacular. Just let me know if you have any allergies. Okay, see you around.;Gus
+Emily's been working here for a while now. I don't know what I'd do without her! I would hate to have to clean all those pots by myself.;Gus
+Pam and Clint come into the saloon almost every night. I'd probably go out of business if they stopped coming. So make sure you don't drive them away!;Gus
+Clint comes into the saloon almost every night and sits at the same table by himself. Sometimes he glances over at Emily with a nervous look. Hmmm. Maybe that's why he comes in so often. Oh well, that means more business for me.;Gus
+Pam comes into the saloon almost every night and drinks way too much beer. Ah well, I shouldn't pass judgment on my customers.;Gus
+Hi there, <PLAYER>. If you're ever thirsty, the Saloon is the place to be.;Gus
+I don't actually drink very much myself. I'm mainly doing this to make a living. Although I do enjoy a taste of the Stardew Valley vintage from time to time.;Gus
+Are you much of a chef, <PLAYER>? If you have a kitchen and some recipes, you can cook some useful dishes. Home-cooked meals make great gifts, as well.;Gus
+Business has been pretty steady thanks to my regular customers.;Gus
+Oh... you're that new farmer boy, aren't you?;Haley
+Oh... you want to dance with me , do you?;Haley
+... You got this for me? Wow, <PLAYER>, this is the best birthday gift ever!;Haley
+Awwwww, you got me a birthday gift? Well, I really like it!;Haley
+Gross! You got me this on my birthday? What's wrong with you!;Haley
+For me? Ooooh... I can't decide if I should drink it, or use it as shampoo...;Haley
+Is that... coconut? I love this! Mmm.... coconuts remind me of warm, sunny days at the beach. When I hold one, I feel like I'm there... even when it's raining.;Haley
+Wow, they're beautiful! These are my absolute favorite.;Haley
+I don't think I've ever set foot in this place. It's actually kind of cozy! Oh, I'm not worried about the rain. I'm sure it'll pass.;Haley
+Wake me up when there's a pink rain... Or better yet, a pink cake rain!;Haley
+It was fun taking pictures, but I had to shower for two hours to get the mud out of my hair!;Haley
+Wow, looks like someone's read the latest issue of 'Super Fashion Girl'... muddy boots are totally in right now!;Haley
+My parents sent another postcard... I guess they're in the middle of a cruise down the Gem Sea Coast. They said me and Emily could join them at any point, but let's be honest... who wants to go on vacation with their parents?;Haley
+The other day, Emily baked something called a 'squirrel loaf'... How'd it taste? Um... well, if my tail was a little bushier I probably would've loved it.;Haley
+Did you see last night's episode of 'Tears'? sob ... I can't believe Kyle and Amy broke up...;Haley
+My name means 'field of hay', but I like to imagine it as a meadow full of sunflowers...;Haley
+I was waiting for you to ask me that.;Haley
+You look really handsome today.;Haley
+Well, I guess this makes you my 'brother-in-law', huh? That's funny to think about.;Haley
+Things didn't work out between you and Emily, huh? Well... It happens. It is kind of awkward that you're both still living in this tiny town, though.;Haley
+So, what's the idea here... am I just going to be your 'girlfriend' forever...?;Haley
+Is Emily settling into the farm life? You know, I was starting to wonder if Emily would ever get married...;Haley
+I saw a dolphin on the way in!;Haley
+There's something about the sun that just makes me feel good.;Haley
+It's strange the places life can take you... who knows what other lands are out there?;Haley
+Wow, look at how the sunlight sparkles on the horizon! I bet the sunsets here would make for an amazing photo. I have to remember to bring my camera next time.;Haley
+Getting the perfect tan is an art form... It's kind of like toasting a marshmallow... you have to rotate the body just right.;Haley
+Laying out on a tropical beach. sigh I think I've finally found my place in the universe. But I guess Stardew Valley's not such a bad place either!;Haley
+Emily's friend had this swimsuit shipped to me from all the way out in Zuzu City. City fashion is on a whole different level, isn't it?;Haley
+I painted my toenails blue to match the sea. What do you think? wiggle wiggle;Haley
+Gus made a cute, fruity drink just for me!;Haley
+This is such a perfect beach. I'd stay here forever if I could!;Haley
+sigh ... My days of being Flower Queen are over... so it's a bittersweet dance for me. But, yes...;Haley
+I'll be fine on my own... Thanks for the memories. Why are you still clinging to me? Go live your own life.;Haley
+Agh! Oh... hmph. Beat me to it, huh? Alright...;Haley
+I've got nothing to say to you!;Haley
+Alright, alright... I'm done giving you the cold shoulder. Just promise me you'll never, ever lie to me again.;Haley
+This town is so small. It sucks. I have to drive, like, twenty miles to buy any decent clothes. That's why I usually just order online. What?;Haley
+The only thing I like about this town is the beach.;Haley
+I've decided I am going to organize my clothes today. I'll have to throw out all of last year's styles to make room for the new ones! sigh I could really go for a cupcake right now. Do you need something?;Haley
+Don't you get tired of running around on that farm all day, or whatever it is you do? I couldn't stand getting all dirty like that. You probably get a nice tan, though.;Haley
+I'm feeling an urge to go shopping. Ugh! I wish there was a mall here.;Haley
+My sister is so weird. Sometimes I wonder if we're actually related.;Haley
+There are only two things I like about this town. One of them is the beach. The other one is a secret! giggle;Haley
+I wonder if any nice shells washed up on the beach this morning? This isn't the best time of year for shells, though.;Haley
+I'm glad that the flowers are starting to bloom. Those pink ones smell so good.;Haley
+I need to find some kind of hobby other than shopping. I've decided I want to expand my horizons. Maybe I should learn to play the mini-harp?;Haley
+I used to complain about this town being so small, but I've grown to like it. If it was much bigger it wouldn't feel like a community.;Haley
+I cooked dinner last night. It actually turned out okay! But I did make a huge mess in the kitchen...;Haley
+I'm thinking about donating a bunch of clothes... I must have a hundred pairs of shoes.;Haley
+Yesterday I found a seagull with her wing caught in a net. I set her free, of course. She looked so helpless, the poor thing. You know, I should probably start exercising more... this youthful metabolism won't last forever.;Haley
+<PLAYER>, can we go on a camping trip someday? The most exotic place I've been is Zuzu city.;Haley
+NO!! I'm not going to do it! You guys are both stupid!;Haley
+She should take the 'high road'? What's that supposed to mean? You think I'm being a baby, huh?;Haley
+Oh... you aren't? Hmmm... Well could you try opening this jar for me anyways?;Haley
+Oh! I thought you knew. This room's for developing photos.;Haley
+No, you won't! This bracelet was passed down to me by my great-grandmother!;Haley
+sigh ...Maybe it'll wash up on another shore. I can't bear to think of it at the bottom of the ocean.;Haley
+Him? No! ...Well, I guess I used to like Alex. But not anymore! He's immature.;Haley
+What? That sounds fun, but I wouldn't want to leave Stardew Valley right now...;Haley
+I'm going to get such a nice tan this summer.;Haley
+Ew, you're all dirty.;Haley
+Um... Yes? I didn't hear you, I'm thinking about something else.;Haley
+Do you have baby bunnies on your farm? Bunnies are SO cute! ;Haley
+If you stay in the sun a lot it'll make your hair lighter. It's a good thing to know!;Haley
+I wonder what Alex is doing today... Huh? I didn't notice you standing there.;Haley
+Farming sounds so boring� What do you even do all day?;Haley
+Hey, you better not be snooping around in my room anymore! But I guess it could get you in pretty good shape.;Haley
+Hmm... sounds like a lot of work.;Haley
+What? You'd better not be doing that!;Haley
+Emily usually cooks dinner... But she makes weird stuff like quinoa.;Haley
+Do you wear those clothes every day?;Haley
+Hey I found this new tool to open jars with. So you won't need to help me anymore. There's not a cloud in the sky today.;Haley
+It's so hot today! I wish I had some ice cream.;Haley
+I spent 3 hours practicing my signature today. I guess that's pretty silly, huh?;Haley
+If you stay in the sun a lot it will make your hair lighter... But you don't want to get sunburnt, either. I just want a healthy amount of sunlight. Everything in moderation, right?;Haley
+I should start reading some books instead of magazines. It's good to learn things, isn't it? I only ever look at the pictures.;Haley
+I talked to my sister for a while last night. I actually enjoyed it quite a bit. You know, Emily's not actually that weird. I guess we have more in common than I used to think. Don't tell her I said so.;Haley
+It seems like a good day to go swimming.;Haley
+I guess living in the country isn't so bad. If I lived in the city I might start to miss all the trees. <PLAYER>, I think I've been hanging out with you too much.;Haley
+Oh, hi <PLAYER>! Do you like kids?;Haley
+It's gross when all these fallen leaves get slimy.;Haley
+Hmm... Something smells weird... You've been working on the farm, huh?;Haley
+I feel the urge to go shopping. sigh;Haley
+Is that a grass stain on your knee? Sorry, I'm allergic to grass.;Haley
+It's too cold to go to the beach anymore. sigh ...what do you think I should do today?;Haley
+I'll probably take a nap later.;Haley
+Hmm... That's a pretty good idea, actually.;Haley
+Get away from me, you stupid jerk.;Haley
+I've never been to the forest. It's muddy and I could get a blister on my foot.;Haley
+Something smells earthy... Oh, right. You work on a farm.;Haley
+You look like you've been doing a lot of hard work. Why don't you take the rest of the day off?;Haley
+If you're ever bored, come say hi.;Haley
+Shopping just doesn't sound as fun as it used to. What's happening to me? Hmm...;Haley
+Did I ever tell you I found another one of my great-grandma's bracelets in the attic? See this bracelet? It belonged to my great-grandmother. She was a really interesting lady.;Haley
+I've never been to the forest, you know. It might be interesting to explore it some day.;Haley
+I'm starting to like the smell of dirt. Isn't that weird? I think it's because you're around all the time.;Haley
+I want to move somewhere warmer.;Haley
+Hmm. What brand is your shirt? Oh. Um, never mind.;Haley
+I wish someone would bring me a peppermint coffee. Don't even bother, I know you won't be able to make one.;Haley
+I'm going to start writing a list of all the clothes I need for next spring.;Haley
+Do you have a pony on your farm? Oh. That's disappointing...;Haley
+The only good thing about winter is that I get to sleep more. I like at least 10 hours of sleep every night.;Haley
+Will you say hi to Alex for me? Thanks.;Haley
+Your name is <PLAYER>, right? I keep forgetting.;Haley
+I think I'm starting to realize that clothes aren't the most important thing. Right? I still like clothes, though. I mean, what's wrong with expressing yourself a little?;Haley
+Maybe I'll donate some of my skirts to a charity this spring. I mean, I do have over 1,000 skirts. ...what?;Haley
+Um, if you see Alex could you tell him I'm busy today? Thanks.;Haley
+I wish it were warmer here in Stardew Valley. I could move somewhere else, but...;Haley
+Do you think you'll ever have a pony on your farm?;Haley
+I'm glad you have less work to do in the winter. It must be nice to have a break, huh?;Haley
+<PLAYER>, you look rugged today. I like it.;Haley
+It's a pleasure to meet you. I'm the local doctor. I perform regular check-ups and medical procedures for all the residents of Pelican Town. It's rewarding work. I hope you'll find your own work equally rewarding, in time.;Harvey
+Oh! Thank you very much! I didn't expect anyone to know it was my birthday today.;Harvey
+Oh! It's a... well, it's something. I didn't expect anyone to know it was my birthday today.;Harvey
+You're giving me this? I'm honored. Thank you!;Harvey
+Oh! You... want to dance with me? ...Yes. I'd love to!;Harvey
+I brought some medical supplies, just in case. Be careful out there!;Harvey
+The first time this happened, we were all worried that the rain might be dangerous. But what if it's actually healthy for the body? Who knows... The plants seem to respond well to it, at least.;Harvey
+It's for me? Thank you! This will be fantastic drizzled on some pasta.;Harvey
+For me? What a gift! This looks fantastic! You know, on a physiological level, I wouldn't advise the consumption of alcohol... But there's no denying the benefit of kicking back and enjoying a glass of fine Stardew Valley wine!;Harvey
+I've been working on an 'animal checklist' for a few years now. I want to observe every species living in the valley!;Harvey
+I tried making my own pickles one time... Um, let me put it this way... The frogs we dissected in high school smelled like roses in comparison...;Harvey
+As a small town doctor, you can't help but form a personal connection with your patients. That can be both good and bad...;Harvey
+Well now that you know my secret, I might as well give you an update... I was significantly less winded after aerobics class this week!;Harvey
+I got such a good workout during aerobics class, I think I can treat myself to some pasta primavera tonight...;Harvey
+I don't know why I was ashamed about being in dance aerobics class. My legs have become so strong, I should feel as bold as a lion!;Harvey
+I've never told anyone about my airplane hobby before... I guess I just feel comfortable talking to you.;Harvey
+As a doctor, you can only do so much to help your patients... I've made the mistake of getting too emotionally involved in the past... I won't let that happen again...;Harvey
+I've been thinking about you so much, I can hardly do my work.;Harvey
+Maru's been unusually cheerful lately... I hear you may have something to do with that.;Harvey
+Congratulations on your marriage, <PLAYER>. Maru's already informed me that she'll continue to work at the clinic, which is a relief!;Harvey
+Sunscreen? Check... Sun hat? Check...;Harvey
+If you get burnt, a little aloe vera cream might help soothe the skin.;Harvey
+I combed the whole beach for glass. We wouldn't want anyone to get a cut on their foot!;Harvey
+It's nice to have an opportunity to relax now and then. The trouble is actually letting go of your worries...;Harvey
+It's important to stay hydrated in this heat. I'm going to ask Gus if he has any coconut water!;Harvey
+I'll let you in on a little secret... I can only swim with floaties.;Harvey
+I've been looking forward to it all afternoon! You look radiant in the fresh spring air, my love.;Harvey
+Oh! I'm sorry... I, er... have plans to ask someone else.;Harvey
+I... I can't look at you. Please spare me any more pain.;Harvey
+...Y...You...? gulp ... I... I think I'm going to be sick...;Harvey
+I'm busy with my work. I can't speak to you.;Harvey
+I want you back in my life, <PLAYER>. I'm willing to forget what happened.;Harvey
+We sell a few over-the-counter medicines at the clinic... feel free to stop by if you're feeling exhausted. I know that being a farmer is pretty tiring work... don't overdo it!;Harvey
+I feel responsible for the health of this whole community... it's kind of stressful. It's a pretty small community, and I'm fortunate to be able to build a good relationship with my patients.;Harvey
+I'd like to stay in Stardew Valley for a while. I'm finally starting to feel involved in the community. It feels good to know that you're useful.;Harvey
+Oh! Hi, <PLAYER>. Do you... have any medical questions?;Harvey
+Feel free to stop by my office if you're ever feeling ill. You're young, though. You'll probably stay healthy without trying.;Harvey
+Don't overwork yourself, <PLAYER>. Doctor's orders! If our farmers get sick, where will we get our food? We need you! ;Harvey
+I'd like to get to know you better, <PLAYER>. Let's put aside our doctor-patient relationship.;Harvey
+It's nice to have a friend in town.;Harvey
+You have a really healthy glow. That's good.;Harvey
+Remember to cover your mouth when you sneeze. Then make sure to wash your hands.;Harvey
+As a doctor, I'm well aware of the benefits of home-cooked food. But I'm so busy, I have a hard time bringing myself to cook. It's a lot of work, being a doctor. I don't eat as well as I should. If I didn't live alone I think it would be easier.;Harvey
+Nutrition is important, so make sure and eat well. Try to increase your vegetable intake! Home-cooked meals are best. Do you cook?;Harvey
+It's a beautiful day, isn't it? I wish I had less work to do.;Harvey
+If you want to hang out in my apartment, that's okay with me. I live above the clinic.;Harvey
+Stop by my place if you ever need a confidential check-up.;Harvey
+Can I do anything for you? Take care.;Harvey
+I came here because I liked the small-town atmosphere, and the potential for a holistic approach to patient care. I've grown to really love it.;Harvey
+Sigh ... I'm starting to feel kind of old... The older you get the more memories you're burdened with. It can be overwhelming. I think I'll stay younger with you around.;Harvey
+Hmm... I'm struggling to make ends meet. I don't have enough patients. I guess I should try to get patients from the neighboring towns...;Harvey
+Imagine having a patient's life completely in your hands, and failing to keep them alive... that's something that will haunt you forever.;Harvey
+Thanks, <PLAYER>. You see, George, I'm just trying to help you.;Harvey
+Sigh Look, George. If you want to behave like this I'm going to have to tell your wife. I don't think she'll be happy.;Harvey
+Oh! Hospitals make you nervous, huh? Just try and relax. I'm here to help.;Harvey
+It's okay to get a moderate amount of sunlight. Just don't get burnt. Okay. Take care! Stay healthy.;Harvey
+Exercise is important, but don't get too exhausted, or you might end up at my clinic! Make sure and listen to your body.;Harvey
+People don't get sick much during the summer. That's nice, but it also means less business for me.;Harvey
+Have you been going into the mines, <PLAYER>? It's a dangerous place. I recommend against it.;Harvey
+When you eat certain foods you'll perform better. Eating a 'Farmer's Lunch' will give you the nutrition you need to water and harvest crops better!;Harvey
+It's flu season, so you'd better be extra careful. Make sure to wash your hands often.;Harvey
+I hate to say this, but I do make a lot more money during flu season. I guess if people stopped getting sick, I'd be out of business. Don't get the wrong idea! I want people to be healthy... really!;Harvey
+Wow, a birthday present! I love it!;Jas
+I hope the rain doesn't turn my dress green...;Jas
+A birthday present? Oh... I don't like this.;Jas
+Are you sure? It smells so good! I'll only drink a thimbleful a day.;Jas
+Wow, it's so pretty! And there's a fairy inside? giggle ... I can hear her singing. She'll be my secret friend.;Jas
+Oh, a little dolly! She's kind of dirty, but I'll give her a bath. Thanks!;Jas
+Oh, I like him! He'll fit right in with my other friends.;Jas
+Umm... I'm not supposed to drink this. Isn't it for adults?;Jas
+Oh... Aunt Marnie doesn't let me drink coffee yet...;Jas
+I see a crab!;Jas
+I've got sand in my shoe...;Jas
+I'm collecting seashells... I'm gonna make a necklace!;Jas
+Aunt Marnie told me not to get burnt... But when I walk fast, it feels cool, so I think I'll be okay.;Jas
+Aunt Marnie says I'm not old enough to see a movie by myself...;Jas
+I hope Shane will still come and see me...;Jas
+sigh ... I hope I can get married some day.;Jas
+Um... Have you ever seen a fairy? Wow! I wanna see one, too;Jas
+...Is Shane going to be okay?;Jas
+I had a nightmare that the house was on fire...;Jas
+giggle ...look, I've got real bunny-jewel slippers!;Jas
+Oh...Are you looking for Aunt Marnie?;Jas
+You can play with my dolls if you want to. Just make sure to brush their hair when you're done.;Jas
+Aunt Marnie won't let me go out after 6 o'clock. It's not fair! I'm not afraid of the dark anymore!;Jas
+I'm not supposed to talk to strangers...;Jas
+Miss Penny is teaching me how to write in cursive. She has such pretty handwriting.;Jas
+... I don't know you.;Jas
+Do you like my bow?;Jas
+I like living on the farm... my favorite is the fresh ice cream from our cows' milk.;Jas
+You smell like grass.;Jas
+Sometimes I look for crawdads in the river. Don't tell Aunt Marnie... but I fed one to a cow once. She loved it.;Jas
+You're always so nice to me, <PLAYER>. Grown-ups are usually mean and boring, but you're different.;Jas
+Um... d...do you want something? I'm not supposed to talk to strangers.;Jas
+Shane's gone a lot, and Aunt Marnie is busy all the time... so I have to invent my own games. I'm thankful for all my toys.;Jas
+Did you know Shane's my godfather? He was a friend of my parents. ...they died. sniff;Jas
+Oh! You aren't exactly how I imagined... but that's okay! It's a quiet little town, so it's very exciting when someone new moves in! Having a farmer around could really change things.;Jodi
+Ah... You sure you want to give away such a precious item? Well, thank you!;Jodi
+The view is breathtaking! Deep blue as far as the eye can see...;Jodi
+We all love it when Gus shows up.;Jodi
+I heard about you and Sam. Hopefully he won't cause you any trouble!;Jodi
+I'm so happy for you and Sam. I think you two are the perfect match! I never thought I'd say it... but I'll miss those awful noises from Sam's guitar...;Jodi
+Are you alright? We've never seen anything like this before... I wish my husband was here. He'd know exactly what to do.;Jodi
+Vincent got up early to go play in the rain. I just hope they're right about it being completely harmless!;Jodi
+Thanks again for bringing that fish! It was some of the best we've had in a while.;Jodi
+whisper Did you hear about Shane? That's so terrible... I hope he's doing okay.;Jodi
+Wow... I heard what you did for Pam... Boy, I wonder what she did to get in your good graces like that? ...I wish I had that kind of money to throw around...;Jodi
+Did you see Pam's new house? Boy... I wish I had that kind of money... I'd love to get the kitchen remodeled, at least...;Jodi
+Poor Clint... they say he's head-over-heels for Emily, but I've heard she only thinks of him as a friend�;Jodi
+I've been doing aerobics class with the ladies every week... Recently, someone new joined the class, but he made us all swear we wouldn't reveal his identity!;Jodi
+Penny's a sweet girl. She's been Vincent's teacher for a while now, and he really loves her.;Jodi
+They say the bus is back up and running. I'm sure Pam is happy about that... I heard she'd been drinking more after losing her job... so hopefully this helps.;Jodi
+What's this? It looks like one of Kent's buddies from the war. Maybe you should show it to him?;Jodi
+Maintaining a household is difficult work... but somebody has to do it. Yes?;Jodi
+It's nice to have a family... but I'd be lying if I said I never miss the freedom of being young...;Jodi
+My husband Kent is a soldier, working overseas. That's why he's not here right now. I know he'll come back safe once his tour is over!! Need something?;Jodi
+Exercise is important for staying healthy. I always make sure to set aside some time for it. As a parent, I don't have much time to devote to myself. So I try and make every minute count.;Jodi
+I wish I could spend more time outside, but there's so much work to do. I started wearing rubber gloves to keep my hands soft. The older you get, the more work you have to do to stay healthy. Ok, bye!;Jodi
+What did you say? It's <TIME> already? My goodness! I still have tons of work to do!;Jodi
+Where does all the time go? sigh I remember when Sam was a little baby wearing diapers. He was such a cute little kid. Don't tell him I said that. You must really like talking to me, huh?;Jodi
+So you're a farmer, huh? Interesting. I've never met a farmer before. Maybe some day you can teach me how to grow vegetables.;Jodi
+The food at JojaMart might not be the healthiest for my family, but with such low prices you'd be crazy to shop anywhere else! ;Jodi
+I'm so glad you're trying to improve that old farm. Once you've got everything in order, it's going to add a lot to our community! It's going to be a lot of work for you, though. Bye.;Jodi
+Yesterday I visited Yoba's Altar to give thanks for my husband's safe return. I'm trying to stay positive, but it's hard sometimes.;Jodi
+I'm taking a break from house chores today. I'm taking the day off. If I don't spend any time outside, I'll go crazy! Plus I don't want my legs to turn soft.;Jodi
+I had a dream that I had complete freedom, no obligation to anyone but myself... ...Then I woke up and realized I had a full day of house chores ahead of me... But I don't regret having a family, <PLAYER>... Sam and Vincent mean everything to me!;Jodi
+Summer is nice, but it also means more work. I'll have to keep the garden in shape on top of working in the house!;Jodi
+How's my little Sammy doing on the old farm?;Jodi
+Housework is frustrating because you're never finished. The day after I sweep the floor, it's dirty again.;Jodi
+What in the world am I going to make for dinner? sigh ...;Jodi
+Maybe I've spoiled the boys a bit. They've never had to do any chores. I guess I want them to be able to enjoy their childhood while they still can. Well, Sam's basically a man now, though. I hope he hasn't grown up to be lazy. That won't do. Being a parent isn't very easy.;Jodi
+Is Sam just bumming around the farm, or is he contributing?;Jodi
+Caroline and I are good friends. I tell her everything.;Jodi
+When you become a parent, you have to leave your old lifestyle behind. Maybe that's not entirely true... but if you're not careful, it will happen.;Jodi
+I loved playing in the snow when I was a kid. The great thing about being a mother is that I get to enjoy it again.;Jodi
+Hi. Do you get tired of people asking about your farm all the time? I think I would.;Jodi
+sigh ... Sometimes I dream about life on my own... without a family to look after. Is that horrible? Don't tell anyone.;Jodi
+What do I do for fun? Hmm... That's a good question. I like sleeping.;Jodi
+Sam makes a lot of noise with that music of his. I hope it doesn't disturb the neighbors.;Jodi
+How's Sam doing? I have to admit, there's less cooking and cleaning to do since he moved out...;Jodi
+We kind of rushed into our marriage, with Kent being a soldier and all... I hardly got a taste of youth and freedom, and now it's too late to ever go back. I can't believe I'm saying this to you.;Jodi
+It would be nice to take a vacation once in a while...;Jodi
+If I was rich the first thing I'd do would be to hire a maid and a chef. My goodness, that would be nice!;Jodi
+I should count my blessings more often. We have a good life here. Some people don't have enough food to eat.;Jodi
+Hello, farmer. I've been gone so long... I feel like a stranger.;Kent
+For me? Well, that's unexpected... But I won't say no. Thanks!;Kent
+Well, I can't read Dwarvish, but it's an interesting collector's item. Thank you.;Kent
+Heh... The rain is a blessing. A reminder that there will always be forces greater than anything conceived of by man. That's actually a comfort, <PLAYER>.;Kent
+Standing on this beach... I feel a rare sense of peace. The ocean is a very special thing.;Kent
+This is a great place to bond with the family. I really appreciate you fixing it up.;Kent
+It makes a father proud to see his boy become a man. You have my heartfelt blessing.;Kent
+The boat ride was fun. I even got to use some of my nautical training to help navigate.;Kent
+My apologies for the other day. I don't want any special treatment. I just want to be a normal person, like everyone else.;Kent
+Erm... okay. Thank you. I'll meet you there.;Kent
+...! ...Where did you find this? ... This man... We fought together in the war, <PLAYER>. He was like a brother to me. But... only one of us made it back. He never lost his courage, though. Even in the face of death. A true hero... I'll never forget him. Thanks for bringing me this... It means a lot to me. Here, take this. It's been sitting in the back of our cupboard for years, and we've never used it.;Kent
+Yep... It's been tightly sealed, so the flavor should still be really potent.;Kent
+I don't know if I'll ever get used to being back home. The peacefulness of the town feels like a mask. That's probably just me though.;Kent
+Things haven't changed much since I've been gone. Well, except that you're here.;Kent
+Things haven't changed much since I've been gone. Except that we have a new farmer in town.;Kent
+I feel like I'm settling back in pretty well. It's nice to have a friend in town.;Kent
+...I lost a lot of friends in battle. What's sad is that I can't even remember their faces. ...;Kent
+You've been here a while now, haven't you? Your farm must be thriving.;Kent
+To be alive is a special gift, and it comes and goes in the blink of an eye... never lose sight of that.;Kent
+I've been up since 4 o'clock... sometimes I wonder if I'll ever get back into a normal routine. Sam has really grown up since I left. He's a man now. I wish I could've been there for him.;Kent
+Are you friends with my son? You probably know him better than I do...;Kent
+Hey, <PLAYER>. Are you enjoying the married lifestyle?;Kent
+It looks like a decent day for fishing, doesn't it?;Kent
+I oughta take up fishing. It would give me something to do, keep my hands busy.;Kent
+Jodi tells me you've really become a part of this community. I'm glad to hear it.;Kent
+I have trouble sleeping, so please excuse me if I seem tired. I'm sure you can understand.;Kent
+I don't know what to do with myself now that I'm back. I'm used to having more structure in my day.;Kent
+You've got a routine to keep you busy, huh? That's a good thing. Too much freedom can be stifling. We all benefit from having some structure in our lives.;Kent
+You keep quiet! Stop trying to turn me against my wife.;Kent
+Don't lie to me! I hate lies more than anything!;Kent
+I wanted to give you this as a 'thank you' for letting me live here with you.;Krobus
+...Have you encountered others like me, in the mines? I'm sorry if they were hostile towards you. You see, we've learned to fear humans... there have been too many... unpleasant encounters. You haven't... slain any of my friends, have you?;Krobus
+Oh, a gift to remember the day that I hatched! Thank you!;Krobus
+The flow of water has been unusually strong today... what's going on up there?;Krobus
+Oh... is this what you call a 'birthday present'? I don't understand humans...;Krobus
+Is that Stardrop Tea? Thank you! One drop makes my room smell good for the whole day.;Krobus
+Hey, look! All my friends are in here! Wait a minute, is that me...?;Krobus
+A gift of lovely perfume? Oh, <PLAYER>... I'm blushing, but you probably can't notice...;Krobus
+This smells delicious! I'll eat half, and use the other half to moisturize my body...;Krobus
+A 1000 year war between my people and the dwarves recently ended. There's still a lot of resentment on both sides.;Krobus
+Please keep my location a secret. If the dwarves knew of me, they'd surely send an assassin.;Krobus
+I try not to make too much noise down here. Imagine if the townspeople heard my weird sounds coming from the sewer drain...;Krobus
+...I'm a little bruised, but I'm okay.;Krobus
+Oh, it's an arctic shard! That's a very rare magical item. I'd bring that to the Wizard if I were you!;Krobus
+I find things here and I sell them. Care to buy anything?;Krobus
+If you put your ear up to the right pipe, you can hear into any house from down here. I wouldn't eavesdrop, though!;Krobus
+Please don't tell anyone about me. Humans tend to destroy things they can't understand.;Krobus
+Please don't tell anyone about me. Most humans are frightened by us shadow people... maybe for good reason. But regardless, if the townspeople knew about me, I might not be safe here.;Krobus
+I am too sensitive to sunlight to go out on most days. The conditions in here are perfect. Care to buy anything?;Krobus
+Please, don't be alarmed. I am different than the others. I've spent a lot of time observing humans. I know you like to shop. Care to see my wares?;Krobus
+Back again? I suppose it does fit into my theory of human behavior.;Krobus
+'Krobus' means 'bridge-crosser' in our language.;Krobus
+On Fridays I stay silent as a sign of devotion to Yoba.;Krobus
+I'd love to see more of the outside world, but it wouldn't be safe for me.;Krobus
+I'll never know what it's like to be human... but talking to you, I feel like I understand it a little.;Krobus
+...Sorry, I'm still a little wary of humans. My shop is still open to you, however.;Krobus
+You're not like the other humans, are you?;Krobus
+You've opened my eyes to the world of humans. I appreciate that.;Krobus
+Hello, it's nice to meet you. You picked a good time to move here... The spring is lovely.;Leah
+Oh, wow... I had a sip of this stuff when I was a kid, and I've never forgotten the taste.;Leah
+Hey, thanks a bunch, <PLAYER>! This makes my birthday feel special.;Leah
+You want to dance with me? Yes! I was wondering if you'd ask me...;Leah
+I had a feeling you would ask me...;Leah
+Elliott's a real romantic... I'm sure your days will be filled with poetry!;Leah
+I'm happy that Shane's in such a good place. I've never seen him smile so much! With his love for chickens, I'm sure he'll fit right in on <FARM> Farm!;Leah
+Oh! Thank you! This one has an interesting shape...;Leah
+Hey, where'd you find that? sniff ... Mmm... This will be perfect, grated over a fresh pasta. Thank you!;Leah
+Mmm... Back in the city, the veggies never tasted this fresh. I love it!;Leah
+This looks delicious... Thank you! There's nothing quite like a fresh, wild salad. Oh, wow, <PLAYER>! This is exactly what I was craving! How'd you know?;Leah
+Please don't feel like you have to spend all your time with me now that we're dating... I know you're a busy person!;Leah
+I sold enough art at that show to get by for another year... that's all I needed!;Leah
+I'm still fuming about the picnic... what a weirdo! Let's never talk about... that person... ever again.;Leah
+If my art continues to sell online like it has, I'll be able to keep making art forever. That's always been my dream.;Leah
+Don't tell anyone, but Willy let me steer the boat for a few minutes.;Leah
+Well, that was deeply relaxing.;Leah
+I love the weather in Stardew Valley, but seeing a different climate now and then helps put things into perspective.;Leah
+The shape of the waves here is awe inspiring. I wonder if I could ever capture that sort of formlessness in a sculpture.;Leah
+I like the heat as long as I'm keeping perfectly still.;Leah
+Mmm, I bet the trees around here would make for an interesting medium to work with. Though Willy probably wouldn't be happy with me if I dragged a bunch of logs onto his ship!;Leah
+I wonder what kind of things you could forage out here. I think I saw some tasty looking walnuts growing on a bush nearby.;Leah
+Have you tried the Pina Colada? I'm on my second one... hehe.;Leah
+Finally! I was starting to get worried you wanted to dance with someone else!;Leah
+I'll be honest. I don't want to dance with you.;Leah
+I've realized I'm happier by myself. So just leave me be.;Leah
+Ohh... Got it. Hah... I really don't have any luck when it comes to this stuff...;Leah
+I don't want to see you for a while...;Leah
+You caused a lot of grief with what you did... but I'm ready to put the past behind us.;Leah
+I came for the sound of the ocean. Beautiful, isn't it?;Leah
+The trees look nice today, don't they? So why did you become a farmer?;Leah
+So are you making as much money as you hoped? Spending time in nature is very rewarding.;Leah
+Hey... That's pretty much the reason I came here, too! I wanted to connect with nature more...;Leah
+That's very sweet to honor your grandfather like that.;Leah
+Where did all these plants come from? It's a surprise, but I'm not complaining...;Leah
+I guess everyone in town was scared of that strange rain? I had no idea... I just thought it was incredible!;Leah
+Imagine if this rain lasted all summer... The plants would take over the entire valley!;Leah
+Hi. Oh, you want to talk? The landscape around here gives me a lot of ideas. The terrain is almost like a sculpture itself. I don't make art for money. It's just an urge that I have. Farming seems like a very rewarding profession. You get to create delicious food for everyone! You're probably busy. Sorry.;Leah
+Hello, neighbor. We both live outside of town. Does that mean something?;Leah
+This morning I accidentally stepped on a bug. Sometimes I think it's impossible to live without destroying nature in some way.;Leah
+I love being around animals. They're so innocent. Farm animals are wonderful, too.;Leah
+Hi <PLAYER>! I was just thinking about you and your farm. I was imagining how good it must feel to harvest a big bunch of crops. Wouldn't it be nice to have help, though?;Leah
+There's actually a lot of wild food in this area, if you know where to look. I've been having fresh salads almost every day.;Leah
+Foraging is a specialty of mine. Someday I'll make you a fresh salad.;Leah
+It's simpler to be friends with the trees. They don't have much to say.;Leah
+We wouldn't be able to survive without nature. It's good to remember that.;Leah
+Hello, <PLAYER>. It's a nice day, isn't it? Stop by my cabin if you ever need someone to talk to.;Leah
+Sometimes, a strange feeling comes over me. My head becomes totally empty... and I feel completely at peace, even blissful. Hey, now! It has nothing to do with drinking wine!;Leah
+Do you ever wonder what it's like to be a different person? Do our thoughts manifest in a similar way, or is it radically different? Don't mind me, I think about weird things sometimes...;Leah
+I've thought about growing some crops, but I think I'd feel bad having to clear out all the wild plants to make space. Farming sounds so fun. I want to try it, someday.;Leah
+If you hear any banging from inside my hut, it's probably just me working on one of my sculptures.;Leah
+If you ever find any interesting looking driftwood, I could use it. It would be special that you gave it to me.;Leah
+There's nothing wrong with that... I think there's lots of beauty to be found in raw, untouched nature! But as a human, I'm also interested in how we shape and interpret the world around us... I guess what I'm saying is that I'm interested in 'art'.;Leah
+You're right... I just wouldn't have been happy back there. It was better for both of us... to end an unsustainable relationship while we were still young and flexible.;Leah
+That's a little harsh... but, yeah. We had different priorities.;Leah
+Oh yeah? Well maybe I'll just leave, then!;Leah
+sigh ... I guess you're right... it was kind of selfish. But can you really blame me for wanting to be happy? Ugh... I don't know...;Leah
+sigh ... Maybe you're right... Humans are selfish creatures. I don't know. I probably shouldn't dwell on it. We all have to make hard choices from time to time.;Leah
+You can tell it's summer by the sweet smell of nectar wafting through the air!;Leah
+I found some wild fruits this morning! Sorry, I don't have any left. Keep looking, I'm sure you'll find something.;Leah
+The sound of farm animals is great, isn't it?;Leah
+The simple things in life are best: a soft summer breeze, majestic clouds, and a goblet full of Stardew Valley red.;Leah
+I have a good eye for wild fruit. Some day I'll make you a nice fruit salad.;Leah
+You can use a tapper to harvest syrups and other useful liquids directly from trees. It's a slow process, but the result can be quite valuable.;Leah
+I hate to be blunt, but if we don't treat nature with respect, our grandchildren will be doomed. Don't you think so?;Leah
+Do you ever walk out into the forest... Pause... And simply enjoy the beauty of nature? It's easy to forget how incredible this world actually is...;Leah
+Hello, <PLAYER>. Do you ever get lonely on that big farm? Stop by my cabin if you ever need someone to talk to.;Leah
+The flowing water keeps my house a little bit cooler in summer. I can't tolerate heat very well.;Leah
+One Summer, it was so hot during the night that I got out of bed and jumped into the river! I hope no one saw me!;Leah
+What should I do on such a warm, lazy day? sigh Any ideas?;Leah
+How quickly the seasons change! It's shocking.;Leah
+I found some wild mushrooms over the weekend! I had a huge plate of wild mushrooms and eggs for breakfast today... I'm still full. Keep your nose to the ground and you'll find some mushrooms yourself... I'm sure of it.;Leah
+I wonder if I could make a collage out of dried leaves? They're just so colorful... I can't help but think of the potential for art projects.;Leah
+Everyone has a unique and interesting perspective, whether they believe it or not. Don't you think so? Although, some people have opinions that I just can't respect.;Leah
+It's so nice to live by the river. I fall asleep to the soothing sound of water every night.;Leah
+One of these days I'll have to make you a wild mushroom stir-fry.;Leah
+Do you have a kitchen in your house? If you have a kitchen, you can cook all kinds of delicious meals.;Leah
+Decorations accentuate the feeling of the season. I think it's a worthwhile tradition.;Leah
+There's a lot of good places to walk around here. It feels good to stretch your legs.;Leah
+One thing I love about this place is that there are beautiful waterfalls. Do you ever spend time watching them? It's soothing...;Leah
+I love the way everything looks when it's covered with fresh snow.;Leah
+It gets pretty cold in my little cabin. I just snuggle up under a huge blanket, and I'm okay.;Leah
+I saw a snow rabbit early this morning! They're rare, aren't they?;Leah
+I'm having a hard time finding fresh food to eat. I've been eating a lot of bread.;Leah
+Sigh ... It's hard to get fresh food in the winter. I bet your farm is pretty well stocked with food, huh?;Leah
+Time seems to move slower here. When I lived in the city, the year went by so fast.;Leah
+Do you ever get cold in that farmhouse of yours? Stop by my cabin if you want to huddle under one of my quilts and drink cider.;Leah
+Do you ever take a whole day off? It's a refreshing break.;Leah
+You should take a whole day off, now and then. Not only is it good for the soul, but in the long run, you'll actually be more productive!;Leah
+This is a precious egg... A baby! I'm glad you brought it to me, I'll keep it safe.;Leo
+I've learned a lot from my parrot family... But I'm sure there's a lot more out there I don't know.;Leo
+When Sun is angry, it feels good in the jungle. Sun can't find us beneath the leaves. We trick him! Haha...;Leo
+...Sorry... I'm shy today.;Leo
+Be careful in the volcano... the spirits don't like outsiders.;Leo
+Have you met the others like you? I don't talk to them much. I feel safer with my own kind...;Leo
+One time... I was at the top of the volcano at night, and I saw lights far, far away. I wonder if there are other bird families out there?;Leo
+I wonder if I could see your home from the top of the volcano?;Leo
+...I'm sad today. You go on... have fun. Let me be sad.;Leo
+When it rains, strange birds visit. I try to speak to them... but they fly away. I guess they don't want to be friends...;Leo
+I saw a big mango floating in the sky. Linus told me it was a hot-air balloon! I thought only birds could fly...;Leo
+When I first arrived, Linus made me a suit out of leaves, like his... But I looked funny... like a fuzzy mango! I think I'll stick to my feathers.;Leo
+I need to find some place to live... But I have a plan... I'm going to rely on my friends.;Leo
+Wow, tastes just like golden walnuts! This is great!;Leo
+I like this weather... it's so humid. It reminds me of the forest back home.;Leo
+It felt bad at first... to leave the island and set foot in a strange land. But, I got used to it after a while. I'm a very flexible person!;Leo
+I felt like a stranger here, at first. But I've made a lot of friends! For once, I feel like I belong...;Leo
+Miss Penny is doing her best to get me up to speed in my schooling. She's really nice... ...Luckily, I remember some things from my old, old life... before washing up on the island... so it's not as difficult as it could be. Miss Penny says she's impressed with how much I know!;Leo
+I'm sure Jas and Vincent thought I was weird, at first. I accidentally squawked at the school table a few times... that didn't help. It was a big change for everyone. But we're good friends now!;Leo
+Between Linus and Miss Penny, I'm going to learn everything there is to learn! Mayor Lewis says I'm on the fast track to becoming a 'well-rounded citizen'... whatever that means.;Leo
+Have you ever noticed all the interesting birds that live here? If you listen closely, they might tell you something.;Leo
+The birds of this valley don't mind the cold so much. I'm a little jealous of their warm coats...;Leo
+Linus has been teaching me about all the wild food in the valley. There's a lot of good stuff out here! Though... I do kind of miss having golden walnuts for breakfast every morning...;Leo
+Linus taught me how to make a special spice berry rubbed fish... He says if I eat enough of it it'll keep me warm this winter!;Leo
+Uncle Linus says I should 'cook' my fish before eating it... I'm having a hard time getting used to it... so dry...;Leo
+Hey, <PLAYER>. Nice to see you back on the island, just like old times. This is fun! You know, I was scared when you first arrived... I knew my life was about to change.;Leo
+You know, Stardew Valley might be my new home... but the island will always be my home, too. It's nice to be able to come back and visit.;Leo
+Maybe some day, I'll visit Zuzu City... But I'm not even close to being ready, yet. This world is already so strange to me, and I'm very comfortable here right now!;Leo
+The parrots want me to thank you for all the golden walnuts you've found. They're really enjoying them.;Leo
+<PLAYER>? Um... I'm a little embarrassed to ask... But... do you think Jas would ever like a guy like me? Or am I too weird? Hahaha... who cares. I'll figure it out some day.;Leo
+So, how was your first night in the old cottage? Your grandpa used to complain about the rickety old bed. But I think, deep down, he actually loved that house.;Lewis
+I've sampled many teas in my life, but none are quite so rare and special as this. Thanks!;Lewis
+Ah, you remembered the old mayor's birthday. That's very thoughtful of you. And it's a nice gift!;Lewis
+I just got off the phone with the Governor... Apparently this rain is supposed to be completely harmless. Just an unusual phenomenon of nature. Still, some of the townsfolk are panicking, which is never good...;Lewis
+Well, that was quite the excitement for our little town! I'm just relieved that everyone is okay. If this ever happens again, I think we'll all be a little more relaxed. It's not the end of the world!;Lewis
+I guess we can expect these unusual rains every year. I think everyone is a little calmer knowing that it's just part of nature.;Lewis
+Ah, you remembered the old mayor's birthday... what's this? Um... well you know what they say, 'it's the thought that counts'.;Lewis
+Hey, I was doing my nightly rounds, and noticed that your first parsnip was ready. Congratulations!;Lewis
+Those melons have been looking real good over at <FARM> farm! You know, that was one of your grandpa's favorite crops back in the day.;Lewis
+Nothing says 'Fall' quite like a pumpkin, don't you think?;Lewis
+I'm impressed, <PLAYER>. You've really fixed up the place!;Lewis
+Congratulations, <PLAYER>. This is a big step forward for you and <SPOUSE>. You've really grown some roots in this town!;Lewis
+Congratulations once again, <PLAYER>.;Lewis
+The divorce paperwork is all squared away. Sorry that things didn't work out...;Lewis
+Another divorce? C'mon, <PLAYER>... you should be a little more careful who you decide to marry in the future...;Lewis
+All the young people seem happy about the new movie theater. A thriving theater will be good for the local economy. I just hope there's nothing objectionable in those modern pictures!;Lewis
+Everyone aboard! We're heading back to the valley!;Lewis
+Vacation opportunities like this could really expand the local economy! I'd better not stay in the sun too long, or my moustache will dry out.;Lewis
+Hehe... look at this place! I don't know how you did it, but this little vacation spot is the talk of the town.;Lewis
+Hehe... Gus is really in his element.;Lewis
+<PLAYER>, I heard what you did for Pam and Penny... now that's the kind of community support I love to see in Pelican Town!;Lewis
+<PLAYER>, did you hear? Someone paid Robin to build Pam and Penny a house! Robin won't tell me who it was, though. Regardless of who paid, I love seeing this kind of community support in Pelican Town!;Lewis
+sniff ... sniff ... can you smell that? I've been catching hints of fresh crab all day...;Lewis
+sniff ... sniff ... Ah, I sense you've been spending some time in the old Community Center. It's good to know someone's taking care of the place.;Lewis
+I've never felt such an overwhelming pride in our community. You've done Pelican Town a great service, <PLAYER>.;Lewis
+Somehow, that old bridge to the quarry is now repaired. I'm shocked... It must've taken a whole team to fix that thing! Who could've done that?;Lewis
+<PLAYER>, did you hear? It seems the old town bus started working again overnight! That old thing has been dormant for decades... This is truly a great mystery.;Lewis
+So, the old Community Center is gone... I guess times change, and good things come and go... Yet I can't help but feel a sense of loss. I just hope this doesn't mar my record as Mayor.;Lewis
+Congrats on winning that grange display competition. Hehe... Just between you and me, I think Pierre needed a bowl of humble pudding!;Lewis
+It's a good season to start thinking about gardening, even for us non-farmers. The valley is known for producing some of the finest spring flowers!;Lewis
+Just place anything you want to sell in the shipping box next to your house. I'll be interested in seeing what kinds of things you put in there.;Lewis
+At the start of each season you'll have to remove the previous season's dead crops. It's easiest with a scythe.;Lewis
+Have you planted any spring crops yet? The season doesn't last forever, you know.;Lewis
+You're in town early, huh? Finished with your farm chores already? Good morning! I'm just checking on the garden as usual. Hmm... are these aphids? Hmm... I hope these aren't scale bugs...;Lewis
+Hmm... tax collection... mumble ...spring festivities expenses... mumble ... Yes?;Lewis
+Ah... There's nothing like a strong cup of tea in the morning.;Lewis
+The saloon can get pretty lively at night. Sometimes we even put a coin in the jukebox chuckle I like to visit once a week or so to catch up on things, anyway.;Lewis
+Good evening! It's nice to see you socializing with the townsfolk. It's good to stop by once in a while and see what the latest news is.;Lewis
+Good evening, <PLAYER>! I'm glad to see you relaxing at the saloon. It's good for you to take a break now and then. Your grandfather always worked himself too hard... I'll have an extra beer in his honor tonight.;Lewis
+I hope things are going well on <FARM> farm. I'd hate to see it all dried up.;Lewis
+Everyone seems a little happier on the weekends, don't you think? I'm on the job seven days a week, though.;Lewis
+I've been Mayor of Pelican Town for over twenty years! No one ever runs against me when it's time for an election. I like to think that it means I'm doing my job well. I like being Mayor.;Lewis
+sob ... I can't believe you would do that... You're terrible.;Lewis
+Gardening is a hobby of mine, so I'm excited for this warm weather.;Lewis
+Ah, my flowers are growing wonderfully! .3 Good morning! Can you smell the sweet scent of these dew-lilies? It's early but the air is already quite warm. Summer is nice for an old guy like me hehe.;Lewis
+How did I become Mayor in the first place? It's a long story. We're both too busy for stories right now.;Lewis
+The moonlight jellies will be here soon. It's one of my favorite things about the valley.;Lewis
+The saloon gets pretty lively at night. I like to visit once a week or so to catch up on things. And to get a taste of Gus' authentic spaghetti.;Lewis
+Hmm... I wonder if it's anyone's birthday today? I'd better check the calendar by Pierre's shop.;Lewis
+Say, are you growing any blueberries on your farm? I do enjoy a fresh blueberry.;Lewis
+Some of the townsfolk are wary of strangers. You'll have to win their trust before they open up to you. Just be persistent and you'll get through to them.;Lewis
+It was around this time of year when I first became Mayor, over twenty years ago. That makes me feel really old.;Lewis
+If you have any suggestions for the town, just drop a note into my suggestion box. Oh, that's right... I don't have a suggestion box.;Lewis
+Every Wednesday I have buckwheat waffles and coffee for breakfast. It's a little tradition I observe to keep the bachelor life interesting.;Lewis
+Hmmm... these strange little mushrooms have been popping up in my wood chips. I wonder what kind they are? .3 Good morning! We're slowly entering the cold months, so I'm seeing my garden to a gentle end. Ah, you're in town early, huh? Finished with your farm chores already?;Lewis
+The saloon gets pretty lively at night. I like to visit once a week or so to catch up on things.;Lewis
+Do you have any bee houses on your farm? If you place them near flowers you can get different types of honey.;Lewis
+Sometimes you can tell if there's something buried underground. Just keep your eyes open.;Lewis
+Have you tried Gus' 'secret' spaghetti sauce? ...The ingredients may be secret, but the smell isn't. I'll see you later.;Lewis
+Are you doing your best on the farm, <PLAYER>? You'll have to accomplish things in many areas if you want to become a well-rounded citizen.;Lewis
+Well, winter is here! That means the townspeople will be spending most of their days indoors. At least you'll know where to find them!;Lewis
+Another year is almost over! I hope you had a productive year!;Lewis
+I've got to start coming up with a plan for improving our town next year!;Lewis
+I'll still come by every night to collect your produce, even in the snow. You still need to make money in the winter!;Lewis
+I hope things are going well for you on the farm.;Lewis
+I've never been married. I have no family. All I live for is to be Mayor. Is it worth it? ... sigh ...;Lewis
+Hi there, <PLAYER>! How're you doing today? If you see Marnie, tell her I say 'hi'.;Lewis
+A stranger?... Hello. Don't mind me. I just live out here alone.;Linus
+This means a lot to me, <PLAYER>. I'll save it for a quiet day in the tent. Thank you!;Linus
+I don't usually read books... But this one is right up my 'alley'... Thank you!;Linus
+That's a good find! I'm always happy when eating wild food. Thank you.;Linus
+I thought about tying the knot, once. But now, I'm happily married to the wind rustling in the leaves, and the frogs croaking under the silver moon.;Linus
+All these strange trees will be gone tomorrow... it's one of the mysteries of nature.;Linus
+How did the moss harvest go? Hehe... My bed is a lot softer now.;Linus
+I've already gathered enough moss for the entire year. How about you?;Linus
+Let me see... it looks perfect. You're a good student, <PLAYER>.;Linus
+No, thank you. I have no desire for money... In fact, I think it's cursed.;Linus
+Ah, that looks fresh. I'll be eating good tonight!;Linus
+What do you have there? Looks like it would make good bait.;Linus
+My old friend, the glimmering boulder, has moved on. It may not seem like an important event, but to me it's a big change. I'm happy for the old rock to see more of the world, though.;Linus
+Very few people have gone as deep into the mines as you have.;Linus
+The water here comes from glaciers high up in the mountains. I traveled to the source once, just to pay my respects to the water.;Linus
+theater? Hmm... for me, the slow procession of nature is all the drama I need.;Linus
+Oh... me? I usually wouldn't, but... okay. It's nice of you to think about me. I'll head over there in a little while.;Linus
+The crisp air of the wilderness is all I care to know. I live out here by choice.;Linus
+...Have you come to ridicule me? I'm just minding my own business.;Linus
+I don't know you well enough to trust you. Sorry.;Linus
+Please don't destroy my tent. It's happened before.;Linus
+I'm happy by myself, you know. I don't need new friends.;Linus
+It would be nice if the townspeople could accept me for who I am. I like living out here in the open air. That's what they don't understand.;Linus
+You can learn to survive in the wild. I have. I think we all have a hidden urge to return to nature. It's just a little scary to make the leap.;Linus
+I spend a lot of time thinking. If you can fully understand the reasons behind your thoughts, you'll have reached a new level of being.;Linus
+I have to be wary of strangers. Most people don't like a 'wild man'.;Linus
+The people here seem nice, but they avoid me. People are afraid of the unknown.;Linus
+You can learn a lot from trees. Spend time with them and they might tell you their secrets. Go in peace, young one.;Linus
+I have everything I need to survive, and more. Nature plays a wonderful tune if you can only learn to listen.;Linus
+Someone was throwing rocks at my tent last night... I just had to wait it out. I don't like to stay in one place for too long. There's just too much to experience in the world.;Linus
+During all these years I've discovered a few secrets about life. You'll have to find out for yourself.;Linus
+I saw a heron wading gracefully through the morning mist. Such are the treasures of a quiet life. It's so easy to get caught up in the noise of modern life. Your best years will pass you by in a formless blur. That's why you've got to learn to slow down.;Linus
+How have you been, my young friend?;Linus
+Ah... summer. The warm sun heats up the cans for me. Then at night I've got a warm meal to look forward to.;Linus
+I have my own reasons for living alone like this. Some things are best left unsaid... I should do some foraging.;Linus
+Whenever you catch a fish, take a moment to thank it for its sacrifice. Even if you don't believe in the spirit of the fish, it's a good mental exercise. We need to be aware of our effect on the environment.;Linus
+You can find rainbow trout in streams right now.;Linus
+This is an easier time of year for me. I don't have to worry about staying warm. Plus, the fruits of the wild are growing everywhere.;Linus
+I've explored deep into the caves. They hold some hidden secrets. Just be cautious if you go in there.;Linus
+Some joker sprayed paint all over my home during the night... It took hours to scrub it off this morning. I need to start gathering lumber for the winter. Do you have enough lumber to keep your house warm?;Linus
+I don't have many responsibilities, so I spend my time thinking. You have to understand your thoughts before you can control them.;Linus
+I don't like to stay in one place for too long.;Linus
+There's nothing quite like a feast of wild mushrooms in the fall.;Linus
+It can get really cold if you live in a tent.;Linus
+After you've truly unplugged from the modern way of life, you'll start to really enjoy the simple things... like the sun, a ripe fruit, or finding a stick with the exact shape you were looking for.;Linus
+Thanks for not shunning me, <PLAYER>. It takes a lot of wisdom to override your base instincts. I'm a human like everyone else. I just have a different lifestyle.;Linus
+I'm sure you understand why I'm cautious of strangers. But let's put that in the past, okay? You and I are friends now, I think.;Linus
+Winter is tough, but you can still fish and forage to get by. If you have any animals you should keep them inside.;Linus
+The trees go to sleep during the winter. It makes it a bit lonely out here. But I wouldn't deny them their beauty sleep.;Linus
+Thanks for stopping by. I was actually feeling a little lonely this morning. So, have you discovered anything interesting in the mines?;Linus
+Sleeping on the ground is good for my back. It's best to look at the positive side of things.;Linus
+Ah, Mayor Lewis told me you just arrived. I sell livestock and animal care products at my ranch. You should swing by some time.;Marnie
+Thanks, <PLAYER>. I always like to have a copy of the Almanac around. It's pretty accurate!;Marnie
+I think you'll love having a little friend on the farm with you.;Marnie
+I'm sure <PET> will be a little shy at first. It takes a while for an animal to get comfortable in a new setting.;Marnie
+Has little <PET> settled in yet? It's such a nice property for a pet to explore.;Marnie
+I just remembered... it's been about a year since you adopted <PET>. I bet you two have grown really close by now!;Marnie
+Oh, you got a horse? That's so wonderful! Horses are amazing animals.;Marnie
+Shane seems a lot more positive and hopeful, lately. I'm so happy you moved here, and that you two get along so well!;Marnie
+I'm glad Shane's moving to your farm... he can keep the family tradition alive!;Marnie
+Hehe... finally a chance to show off all my hard work at aerobics club.;Marnie
+Hee hee... It's interesting to see everyone in their bathing suits.;Marnie
+I love animals, Mr. <PLAYER>. If you treat yours well I'm sure we'll become good friends!;Marnie
+Animals are so innocent, so sweet. And If I don't look after them, who will? I just hope my chickens aren't too upset when I take their eggs.;Marnie
+You've been here a while, now... how's your farm doing? Taking care of animals is easy as long as you don't forget about them!;Marnie
+Now that I'm here I can finally relax and socialize a bit.;Marnie
+I spilled a whole bucket of milk on the floor today, it was horrible. And such a waste of my poor Belinda's hard work. It's nice of you to show up! It's a little more lively now.;Marnie
+Phew! It sure feels good to kick back after a hard day's work.;Marnie
+I always see Clint here but he won't talk to anyone. He just glances over his shoulder now and then.;Marnie
+Now that Shane's gone, I'm going to have to teach Jas how to care for the chickens. My nephew Shane has been staying at my place the last few months. He helps me out with the chickens, so I'm not complaining.;Marnie
+You're not worried about this rain? That does put me at ease a little... I just hope the cows tucked themselves in...;Marnie
+Well, looks like we all got scared for nothing... The rain was totally harmless!;Marnie
+The animals seems to enjoy this rain... I think it's putting extra nutrients into the grass.;Marnie
+You can use a scythe to cut feed from grass. Or you can buy it from me, of course! I could use the cash... Adios.;Marnie
+Hi. Need something? I might be getting some different kinds of animals in the future. Although, I'll need to make sure there's a market for them, so you'll have to prove your skill with animals first. Just keep treating your animals with respect and good things will happen.;Marnie
+You can catch me at the saloon most nights. Animals are great company, but I need to spend time with people, too.;Marnie
+I wonder if Mr. Lewis gets lonely in that big house of his... It seems so empty in there, doesn't it?;Marnie
+Have you been to that strange tower west of my house? One time I heard this terrible, otherworldly noise coming from there. I would avoid that place if I were you...;Marnie
+Hey there, it's good to see ya! Feel free to visit us any time you please.;Marnie
+Hi <PLAYER>! I'm glad we've become friends. I know I can trust you to take care of my dear animals.;Marnie
+Well, I didn't win... but it was still satisfying to share all my hard work!;Marnie
+That was some strange-looking purple lettuce in your grange display! I swear, it looked just like Mayor Lewis' special und... Oh!... heh... never mind!;Marnie
+Oh! Aren't you the one who just moved in? I've been looking forward to meeting you! You know, with a small town like this, a new face can really alter the community dynamic. It's exciting!;Maru
+Heh... sure! Suddenly, the dance sounds a lot more fun.;Maru
+Hey, what's this, a birthday gift? It's great! Thanks for remembering me!;Maru
+Hey, what's this, a birthday gift? It's... uh, well, it's a gift! Thanks for remembering me, anyway.;Maru
+Mmm... How does it taste so good? This is really a great gift.;Maru
+This will be perfect for powering my latest project! Thanks, <PLAYER>!;Maru
+Ah, pure gold... It's not the strongest material, but it has some really useful properties!;Maru
+Oh, wow! That's a rare element. It's theorized that all the iridium in our world comes from meteorites...;Maru
+Ooh, this is exciting! I'll deconstruct it and see if I can reverse-engineer the technology.;Maru
+Ahh! Be careful with that! Are you touching it with your bare hands? ... Sorry... hehe... With proper handling, this will be perfect for an exotic machine I'm building.;Maru
+Ummm... Don't break the news to my Dad, okay? I'll handle it.;Maru
+Mmm... I'm having a hard time focusing on my work today.;Maru
+I want to go collect samples with my dad, but my Mom is insisting we stay here...;Maru
+This is a rare opportunity to study a unique phenomenon... every moment counts!;Maru
+I should've spent more time on that piece before showing you...;Maru
+How's your hand feeling? Let me see it... Hey, it looks all better!;Maru
+Hey, I guess we're brother and sister now!;Maru
+Hey, congratulations! I remember the first time Penny told me she had feelings for you... She was all shy about it, but I knew right away it was serious... she wouldn't say something like that if she didn't mean it!;Maru
+You always seemed far too healthy to be at the clinic so often... Now it's all starting to make sense...;Maru
+Willy mentioned that you helped fix the boat. Thanks for doing that!;Maru
+I've been noticing a few different starfish species. My favorite are the big purple ones, but they're all very interesting!;Maru
+gasp ... It's a solar eclipse! Oh... never mind. It's just your body blocking the sun.;Maru
+We're so far removed from civilization out here. Imagine the stars you could see at night. Who knows, maybe we'd even see a flying saucer or two!;Maru
+Mmmm... Why are tropical fruits so good?;Maru
+I wonder if Willy would let me bring my telescope here sometime... The sky is so clear here, I bet I could spot Pelican Town out in the horizon. Maybe I'd even see you running around <FARM> Farm!;Maru
+Of course... whisper Then can we go home and snuggle?;Maru
+I'm flattered! But... No.;Maru
+Not good enough for you, huh? Fine...;Maru
+You've got some nerve, speaking to me!;Maru
+You've lost some respect with me, <PLAYER>... but I'm willing to forgive you.;Maru
+Whenever I'm struggling with a technical problem, I always take a walk. It's surprising how much a change of scenery can help.;Maru
+Have you met my mother? She's the town's carpenter.;Maru
+On Tuesdays and Thursdays I work at Harvey's clinic. He says he likes having me around in case his medical equipment goes haywire! Being a farmer must be pretty easy, huh?;Maru
+I plan on spending a lot of time with my telescope this summer.;Maru
+Sometimes, when I'm looking into space with my telescope, I get struck with this eerie feeling... Like we aren't meant to see this. It's silly, I know...;Maru
+Do you know my Dad, Demetrius? He's a scientist. I have a lot of fun helping him out in the laboratory.;Maru
+It's really satisfying when you can repair something yourself, instead of just throwing it away.;Maru
+Some people say 'magic isn't real'... but have they ever stopped and really thought about magnetism?;Maru
+If you think about the human body, it's hard to believe that somehow, it all just 'works'.;Maru
+You know, I was thinking... even with all the technology in the world, nothing beats having a real human friend.;Maru
+Hey, sorry if I seem cranky... I'm a little sore from work yesterday. I had to carry ten boxes of medical files to the attic! I had to sort patient records for four hours straight;Maru
+Hey, you've made another week of progress on the farm! It seems like things are going well for you. Another tiring week on the farm, huh? It must really be a lot of work...;Maru
+Yikes! I'm sorry. Have you ever thought of doing something different...?;Maru
+Stop by our house when you need to upgrade your buildings. It's always nice to have visitors.;Maru
+My mother is a carpenter and my dad is a scientist. I guess it makes sense that I'm into building gadgets.;Maru
+My Dad is very logical, and my Mom is very intuitive. That difference leads to conflict, sometimes...;Maru
+Sometimes my head gets so cluttered with nonsense I can hardly think. Does that ever happen to you? Getting some fresh air usually helps.;Maru
+I'm glad to have a job at the clinic... I feel like I'm helping people in town. Dr. Harvey? He's nice, but not really my type...;Maru
+Oh, <PLAYER>! Is there anything you want to talk to me about?;Maru
+Of course he'll know the difference! This sample is completely ruined!;Maru
+Okay. I'll blame it on you, then.;Maru
+You're right. I should just tell him the truth.;Maru
+I guess you could see it like that... But it's still incredible, isn't it?;Maru
+Oh no? Well, that makes me feel a little better, then. Thanks.;Maru
+Oh no! Aghh... I'm really really sorry.;Maru
+Thanks. That means a lot to me, <PLAYER>. I know gadgets aren't really your kind of thing... but I do appreciate that you're showing interest in what I do. I wish I could return the favor somehow.;Maru
+What? After she said point-blank that she doesn't want to be a slave? How cold...;Maru
+Heh... Yeah I guess he did kinda say that, didn't he? ...Um...;Maru
+... You have a point, I did program her to act human... but her neural net is so complex, I can't be sure consciousness didn't emerge on its own. Furthermore, isn't it a little arrogant to assume that human-like consciousness is the only worthwhile vessel of existence? Computer brains are different than ours, but that doesn't mean we can disregard them.;Maru
+That's horrible! What's gotten into you, <PLAYER>? I'm just going to forget about this conversation... you aren't acting yourself.;Maru
+Hmm... What sort of gadget should I tinker with this summer?;Maru
+Come to my room if you ever want to tinker with my gadgets.;Maru
+Sebastian is my half-brother, if you didn't know already. We look totally different.;Maru
+Sebastian hardly ever talks to me. I wonder if he even likes me at all? It's sad, I want to have a brother but he doesn't act like one.;Maru
+Penny and I have been doing a seasonal book club. It's just the two of us, but it's still fun to discuss what we're reading!;Maru
+Sorry, I was just thinking about some gadgets I'm making. I've decided I'm going to build a robot someday.;Maru
+Really? Are you sure you're not just trying to flatter me?;Maru
+You're right, but I love a challenge!;Maru
+Hey, have any interesting farm stories to tell me? Guess not.;Maru
+I worked at the clinic yesterday. We didn't have a single patient. I guess people get sick less during summer.;Maru
+I bet you're growing some interesting crops right now. Dad says a lot of exotic plants grow in the valley this time of year.;Maru
+Hey, tell me a story about farming. That's really interesting!;Maru
+Is it just me or did it suddenly get really hot? I guess it's just me.;Maru
+Some day I'm going to be a world-class inventor. You're lucky to be friends with me! ...just kidding.;Maru
+How do your hands stay so soft with all that farm work?;Maru
+Have you explored the mines much? I've heard you can find some useful materials there.;Maru
+In other parts of the world it's spring right now. Imagine that!;Maru
+I bet we'll start seeing more people at the clinic this season. I hope everyone stays healthy.;Maru
+All these falling leaves will provide food for mushrooms and other small organisms. Interesting, isn't it?;Maru
+<PLAYER>, I just love how the trees look right now! I know you like them too.;Maru
+If you have any extra metal laying around, I could always use it for one of my projects. But only if you don't need it yourself.;Maru
+I've been feeling especially optimistic lately...;Maru
+You should come visit me at the clinic tomorrow. I work from 10 a.m. to 4 p.m.;Maru
+You have a leaf in your hair. Let me get it for you. ...;Maru
+Did you see the snowman I built?;Maru
+My hair gets so frizzy in this dry weather. sigh;Maru
+I guess winter is a good time to work on my gadgets. I wonder if I could make anything useful for farmers?;Maru
+Hey! What's up? It's a good time to explore the caves. I would go but my Dad would flip.;Maru
+I hope I don't catch the flu this year.;Maru
+Have you built a furnace yet? You can use them to turn raw material like ores into more useful products.;Maru
+I get so sleepy when it's cold...;Maru
+Hey, if any of your farming machines need to be fixed, just let me know.;Maru
+Hey, do you have any ideas for a new gadget I can make? Mhmm. I trust your opinion.;Maru
+You look fresh today. I guess it's that crisp winter air.;Maru
+Hi <PLAYER>! Hey, don't you think my snowman looks kind of like you?;Maru
+<PLAYER>, come closer. I'm cold. My hands are so dry from the winter air. I'm sorry about that.;Maru
+Keep it up, kid. I think you're gonna be a star. Just ask me if you need anything.;Mister Qi
+I knew you'd end up here someday. Of course, I had to make it a little difficult so you'd have an incentive to achieve excellence.;Mister Qi
+I was born an ordinary boy, but I made a decision early in life that I would become extraordinary.;Mister Qi
+Hey. Don't tell anyone about this place, okay? I want to keep it... exclusive. You'd keep a secret for your friend, wouldn't you?;Mister Qi
+Never stop striving to achieve perfection. Settle for nothing less than excellence in everything you do.;Mister Qi
+Stop by whenever you wanna relax.;Mister Qi
+Keep working hard and maybe you'll end up like me someday.;Mister Qi
+That was a very auspicious wedding ceremony, <PLAYER>. ...What? You didn't see me there? Hahaha... Kid... You can't keep a secret from me.;Mister Qi
+There's more to this world than meets the eye, kid. Some day, you might understand. For now, just keep striving for excellence in everything you do.;Mister Qi
+I got to where I am through the force of pure willpower, kid. You see... The mind is a very powerful tool. In fact, it's the most powerful tool in all of reality... and beyond.;Mister Qi
+I'm impressed with your tenacity, kid... it took a lot of walnuts to get this far.;Mister Qi
+Wherever you go... I'm always behind the curtain, making sure the show is interesting. Hahaha... Or... am I?;Mister Qi
+Set your eyes on the furthest star, and let that be your goal. It's always within your power to go one step further than you used to believe.;Mister Qi
+Some things transcend time and space, kid. It's a unique feeling in the heart... a whisper of things beyond.;Mister Qi
+On the left, you'll find a board with my special challenges... custom-tailored for you, kid. On the right, a set of fantastic prizes for those who meet my challenge... Good luck, kid.;Mister Qi
+Hey, kid. Don't be a jerk and we'll get along fine.;Pam
+Well I'll be darned... I never expected to get a gift like this. You made my day, kid!;Pam
+Looks like there's some unique fish swimmin' around. Next time I'll have to bring my pole!;Pam
+Heh... Willy gave us each a piece of his fish jerky on the ride over. Delicious!;Pam
+I'll really miss that girl... You treat my Penny good, now. You hear me?;Pam
+So, you married the Mullner boy, huh? Will you be takin' that smelly old dog with ya, too?;Pam
+It's a sign from the Almighty... We're doomed...;Pam
+Well, I guess the end-times haven't come just yet, kid... Ah well, back to the ol' nine to five...;Pam
+I think these rains are a show of force... a warning. You could say it's from nature, from Yoba, whatever you like... but that's what I think. What I mean is, we can't get too cocky. Or else, instead of a green rain, maybe we'll have a blood-red rain... See what I mean?;Pam
+Penny came home crying... what'd you do to her?;Pam
+Heard you got a fish pond built... Heh, I grew up with a fish pond. You can raise just about anything in one of those... But some of the fancier fish are real fussy!;Pam
+Alright, alright... you beat me fair and square at the fishing event, kid. Hmm... maybe my problem was using that beer-battered mushroom as bait...;Pam
+Now this is comfort! I might even take my shoes off! Better not look too closely at my feet, kid.;Pam
+It's a special occasion, kid... I'm ordering something unique. Make it a double!;Pam
+Wow, would you look at that... It's like something out of the future. Hehe, sorry kid... I can't use that. There's more buttons than I have brain cells!;Pam
+I can't believe you bought me a house! Is there any catch? Sorry, I'm not used to people being so nice...;Pam
+I'm loving the new house. Thanks again, kid!;Pam
+I can't believe someone bought me a house! I wonder if it's some kind of scam? Sorry, I'm not used to people being so nice...;Pam
+Hey, did you hear about the ol' bus? Yep... I'm back on the saddle, kid... hehe. Feels nice.;Pam
+Urghh... my head... You know, I'd eat healthier food if I could afford it. Hey, you probably have a lot of tasty grub growin' on your farm, hm?;Pam
+Nothin' like a sip o' the good stuff to warm these old bones...;Pam
+Ah... Finally. I need a drink. Gus! Gimme the usual. Hey, I'm back! Ooh boy, am I thirsty! I'm parched!;Pam
+Penny! Did you make dinner? Penny! You done your chores? Penny! You home? Hey! I'm home!;Pam
+Hey, kid. I'd buy you a drink if I could afford it!;Pam
+Hey Gus! Gimme another round! hic;Pam
+Sigh ... Hey, kid. You're too young to be moping around in a place like this.;Pam
+gulp ... I've been looking forward to this beverage all afternoon. Heh heh.;Pam
+Oi, Gus! Gimme another pint of your strongest!;Pam
+Hey, you! Don't be snoopin' around the trailer when I'm out. Got it? Sorry, but I gotta be cautious with strangers.;Pam
+I was reading the newspaper this morning but then I got depressed. It's a rotten world, kid. Keep your head screwed on right and you'll make it through in one piece... That's what my Pappy always used to say. Heh heh heh.;Pam
+You don't know how glad I am that the bus is up and running! I feel alive again. Times have been tough lately. I got laid off of my job as a bus driver. If I got that job back I could drive you to Calico Desert.;Pam
+I had one too many caramel porters at the saloon last night... Go away. My house ain't pretty but at least it's by the river. Hey. Penny's my baby girl. Be nice to her or leave her alone, got it?;Pam
+Penny says I spend too much time at the saloon...;Pam
+It's awful lonely at home all by myself...;Pam
+You know, I've been thinking... I wish I had a hobby. Something to do other than hanging around at that saloon every night. You got any ideas? ...Ehh. Maybe I'll play checkers against myself.;Pam
+Each day's just the same as the last... If only I'd been born rich...;Pam
+Make sure your boots are clean before you go stompin' around in my house. It's annoying to clean a mess. You should know that by now.;Pam
+I wish a team of elves would come during the night and tidy up my house. Hahahaha. I could sure go for some parsnips. If my legs weren't so stiff I'd visit the mountains every now and then.;Pam
+Are you friends with Gus? He's a pretty good guy. He even gives me free beer on my birthday.;Pam
+Hey, watch your mouth! No one tells me what to do!;Pam
+I had hopes and dreams like you, once... But look at me now. Life doesn't always turn out like you expect. Ah well. There's no sense dwellin' on what you don't have.;Pam
+Havin' a local farm is good for our town.;Pam
+Maybe if you do real well on your farm it'll boost the local economy. Then everyone will have a job and we'll all be happy, right?;Pam
+Oh, hi there, farmer. I don't often chit-chat with strangers.;Pam
+I always seem to end up at the saloon at some point. It's too close to my house. It ain't healthy.;Pam
+...Yeah? You're a fine lookin' young man. Why're you wastin' your time talkin' to an old girl like me?;Pam
+Don't mind that ol' dog next to the house. He might look at you cross, but he'll never get up from that box in a hunderd years. The old boy's name? It's Dusty.;Pam
+Someone's been complainin' about my yard... sayin' it's a mess. Was it you? Hmmm...;Pam
+If I knew you better, I'd be more inclined to chat.;Pam
+I used to be young, once. Hard to believe, huh? Time sure goes fast when you grow up.;Pam
+Sigh ... I guess I'll be popping a frozen dinner into the microwave tonight.;Pam
+Penny's the only one I've got, so you best treat her kindly.;Pam
+Every day is the same old routine... Do you ever have that problem, farmer?;Pam
+Would you do me a kindness and clean up my house? Just kidding.;Pam
+Don't you have work to do? You should be thankful that you have a job rather than moping around town all the time.;Pam
+I miss my old truck driving job. I used to go all around the valley, meetin' people, listenin' to the radio. Times sure do change.;Pam
+You got any hogs on that farm?;Pam
+Do you want somethin' from me, kid? Best look somewhere else.;Pam
+I just hope Penny has a good life ahead of her. If you ever have kids you'll know how I feel. ...It's so dry in winter. My skin feels like old parchment.;Pam
+Hey. You better be treatin' Penny right on that old farm of yours. She's the only one I got!;Pam
+Hey kid. What're you up to? I'm just passin' time, same as always.;Pam
+It's awful cold. A visit to the saloon should warm me up, don't you think? Heh heh. Why don't you swing by the saloon later for a taste o' the Oh-Be-Joyful?;Pam
+Penny sure does like old books. I think she's a heck of a lot smarter than I ever was. I'm glad.;Pam
+I oughta take up fishin', seein' as how close the river is.;Pam
+Blah... my front yard's a mess.;Pam
+Oh! Is this for me? For... for my birthday? sniff ... Sorry. I'm... just not used to such kindness... ... I love it, though. It's perfect. Thank you so much.;Penny
+For my birthday? That's so nice of you! It's a great gift.;Penny
+For my birthday? That's so kind of you! Let me unwrap it... ...What!? ...Oh, I see. It's a cruel joke. Silly me, I should've known better...;Penny
+Oh... with me? Yes, of course! ...Will you take the lead, though? I'm nervous...;Penny
+<PLAYER>, what do you think... should a person be judged by their actions, or by their intentions? I was just wondering what you thought...;Penny
+Mmm... it's so fragrant. I'm going to love this.;Penny
+Wow... this looks really old. I'll have to ask Gunther about it. Thank you!;Penny
+Thank you! I'm almost finished with a book right now, but I'll read this one next!;Penny
+I've been teaching Jas and Vincent about Dwarvish Archaeology... Jas is curious about the origins of the Dwarves... but Vincent is mainly interested in dwarvish weapons!;Penny
+Vincent has been asking me about slugs for weeks... so I prepared a whole lesson on them. They're actually pretty interesting! Whenever I see a banana slug, I feel like it's a sign of good luck.;Penny
+I'm teaching the kids about art history this week. It can be a little dry for the kids, so I'm having them learn by drawing a picture in the predominant style of each century. It's really been a hit!;Penny
+Oh, pardon this bandage on my hand... I went over to Maru's the other day and she showed me a new gadget of hers. The darn thing charred my palm!;Penny
+I've been reading a book about the ocean. It accounts for most of the area of the world, and yet we hardly know anything about it!;Penny
+Suddenly, the world looks a lot more beautiful, don't you think?;Penny
+I'm trying to finish this book, but I keep getting distracted thinking of you.;Penny
+I never thought my life would be more exciting than a novel, but it sure feels that way right now!;Penny
+That was a beautiful wedding. I know Maru will make you very happy, she's a great person.;Penny
+Hi, <PLAYER>. Maru told me the news... I'm happy for you two!;Penny
+Congratulations, <PLAYER>. Sam's a lot of fun, I'm sure you'll never have a dull moment!;Penny
+I'm a little embarrassed wearing my bathing suit in public. I hope no one's looking...;Penny
+Somehow, the water here looks so different... I can hardly believe this is the same ocean I've been gazing at my entire life.;Penny
+The wind keeps making me lose my page, but it feels really great on the skin.;Penny
+I'm reading a book about a sailor who washes up on a deserted island. I wonder what I would do in that sort of a situation?;Penny
+Hey honey, do you mind standing a little to the left? Thanks, the sun was making it a little bit hard to read!;Penny
+This sand is so soft and fine. It feels great on my toes!;Penny
+I accept! whisper Thanks for going through with the formalities, dear.;Penny
+... I... I see... sniff ... W... We can just be friends, then.;Penny
+What you did for my family is so generous. Thank you!;Penny
+There are some truly wonderful people in the world...;Penny
+I can't bear to see you...;Penny
+You've left a permanent scar on my heart, <PLAYER>. Still... I'm ready to speak to you again.;Penny
+sigh ... My mother definitely has a problem with going to the saloon too much. But it's best not to dwell on bad things, right? ;Penny
+But if I keep trying I think I can get her to change.;Penny
+That's how I feel. I'm just going to focus on making the future better.;Penny
+Maybe you're right. It's better to cope with reality.;Penny
+I'm eager to move out. It's such a burden to be worrying about Mom all the time. I want her to be happy, but I can't stay here forever... you know?;Penny
+I'm tutoring Vincent and Jas today... They're a handful, but it's nice to make a difference in someone's life.;Penny
+Maybe now that my mother has her bus driving job back, I can start saving more money. Things are looking a little brighter. I'm trying to save money from my tutoring job, but it's hard with my mother out of work.;Penny
+Um, <PLAYER>? A farm seems like a good place for children, don't you think? It was just a thought I had.;Penny
+We don't have a school here but I'm doing my best to give Vincent and Jas a proper education. Every child deserves a chance to be successful. Jas is very good at math and reading. Vincent is good at... well, he has an active imagination!;Penny
+You know what? I should take the children on a field trip some time. Maybe to the forest. A guest speaker would be nice... maybe someone familiar with nature? Sorry, I'm just thinking out loud.;Penny
+This strange weather is a good opportunity to get the kids interested in nature!;Penny
+I've lived in Pelican Town my whole life. Can you believe that? I guess there's a lot out there I'll never experience.;Penny
+We're very lucky to have a library in such a small town. When you're lost in a book, it's easy to forget the realities of your life. ...Maybe that's why I like reading so much. ...Sorry. I got carried away there.;Penny
+I wish I could keep a garden, but our yard is such a mess. Maybe I'll live in a place where I can have a garden someday. Maybe I'll live on a farm! giggle;Penny
+If you dig in the dirt you can find some interesting things. One time I found a really old piece of pottery. I brought it to Gunther and he said it was over a thousand years old.;Penny
+I shouldn't complain so much about chores... some people in the world are really struggling, and here I am complaining about doing dishes...;Penny
+Dishes, dishes, dishes. Ugh... If my mother wasn't always nursing a headache from her late nights at the saloon, maybe she could help around the house a little.;Penny
+This is such a small town. You can't avoid meeting everyone. I wonder what it's like to live in the city?;Penny
+Things changed a lot after the JojaMart went up. It's been really bad for Pierre's shop. Mom loves JojaMart, though. The prices are cheap, so she can afford a lot more there than she ever could at Pierre's. Mom's been complaining ever since JojaMart closed down. I think it's good for her, though. Maybe she'll drink less Joja Cola now.;Penny
+Hello. Ummm... The weather's interesting today, don't you think? Sorry...;Penny
+I've been trying to keep our place clean, but it always gets messy again. It's hard to run a household all by yourself.;Penny
+Oh no? Well, I guess that makes sense... you're busy with other things right now.;Penny
+... Me too. I'm glad you feel that way.;Penny
+Yes... the urge to care for something innocent and helpless. It makes sense that we'd feel that.;Penny
+Oh, really? I think you'd make a good parent.;Penny
+hmm... If everyone thought like that, humans would die out.;Penny
+Oh... That's kind of sad... but I guess I can understand your point.;Penny
+Great! Let me just call the children over.;Penny
+Really?... Uh... Well, alright. I guess I'll see you later then.;Penny
+Thank you... I just wish George wasn't so upset... I was only trying to help.;Penny
+Oh... I guess you're right. I'm sorry, Mr. Mullner. It was rude, what I did.;Penny
+I guess you're right... why stress out about something you can't change?;Penny
+You're right, we shouldn't ignore the reality of aging. I guess the sooner we come to terms with our mortality, the more time we can spend really living in the here-and-now.;Penny
+That's nice of you to say... I totally agree with you. We should treat our elders with the same respect we hope to receive ourselves some day.;Penny
+That's a horrible thing to say. Life is a precious thing to waste like that!;Penny
+You really mean it? Thank you! ... it's such a relief to hear that. I've been working so hard on this recipe, and I'm really proud of it.;Penny
+...Oh, <PLAYER>. I thought you did, but I wasn't sure. ...I'll always remember this night.;Penny
+Thanks for fixing the community center back up. I guess you're kind of a town hero. Hello, <PLAYER>. It's nice to see you're doing well.;Penny
+I heard you were responsible for fixing the bus back up. Thank you. My mother used to drive the bus to Calico Desert... but the bus stopped working a few years ago. Mayor says there's not enough money in the town coffers, or else he'd have it fixed.;Penny
+So... do you like to decorate your farm house? It must be nice having your own place to decorate.;Penny
+Hello. Ummm... The mountains look nice today, don't they?;Penny
+Hi... Um, are you doing okay?;Penny
+It's kind of embarrassing to be the only family in town who lives in a trailer. I wish I could live in a nice house, someday.;Penny
+Mom doesn't take very good care of herself... I worry about her.;Penny
+Fall is a sad time of year for me...;Penny
+When I was a little girl, my father abandoned us because he said he felt 'trapped'. We haven't seen him since. I'm sorry to make things uncomfortable for you... It's just that this time of year makes me think of it for some reason. Anyway... How's the farming life going?;Penny
+<PLAYER>! I was just thinking about the future... Where do you picture yourself in ten years? I have a few ideas for myself...;Penny
+Hi <PLAYER>. Your farmhouse must be pretty easy to clean, huh?;Penny
+Are you friends with Linus? He lives in a tent up in the mountains. Everyone ignores him, the poor guy. He's actually really nice... Just a little odd. It must be so cold to live in a tent this time of year.;Penny
+I'm looking forward to seeing what your farm looks like in the spring. It's going to be so fresh and beautiful.;Penny
+I try to stay indoors as much as possible during the winter. I hate being cold.;Penny
+Oh... hi. You want to talk? If you find any interesting artifacts or minerals, Gunther will gladly display them in the museum. He told me he has special gifts for people who donate.;Penny
+On the 25th we'll be having the Feast of the Winter Star. It's one of my favorite events. It's supposed to be a time to show gratitude for all the good things in your life... but for most people it's just a time to relax and exchange gifts.;Penny
+Hey, it's Mr. <PLAYER>, the new farmer! I'm owner of the local general store.;Pierre
+How can I possibly compete with Joja? Their pockets are so deep, they can just sell at a loss until little guys like me are out of business... then they'll jack up their prices!;Pierre
+Wow, I can't believe you'd give this to me. It must've cost you a fortune! Thanks!;Pierre
+Gripping! Fascinating! It's the perfect book for me!;Pierre
+Gahh, no customers! Any more of this 'green' rain, and my ledger will be in the 'red'!;Pierre
+Well, I'm glad things are back to normal. Seems like it was all overblown... just like everything.;Pierre
+Glad to see you, <PLAYER>. What's a little toxic rain when you have seeds to buy? I'm just kidding. The rain seems to be completely harmless.;Pierre
+Abigail's going through a rebellious phase... I get it. She'll wise up when she has to make her first mortgage payment!;Pierre
+I'm determined to keep my shop running until the bitter end. Morris and his cronies will have to drag me out from behind the counter, kicking and screaming!;Pierre
+I'm losing a lot of money being away from the shop... But, you know what? It's worth it.;Pierre
+Mmm... that salty smell of the sea is making me want some sashimi.;Pierre
+Eh, let the kids play in the sun. I'll just sit in this chair.;Pierre
+Drink after drink... Gus must be making a killing!;Pierre
+Well, now that you have dual incomes, I expect you'll be purchasing twice as many seeds, right? Haha... I'm only joking.... ahem ...;Pierre
+This is a dream-come-true! You can't begin to imagine how much stress JojaMart was causing me. For a small town like this, one store is enough! There's too much social pressure for me to overcharge, anyway.;Pierre
+This is a sad day for Pelican Town... The stronger Joja becomes, the closer I move toward bankruptcy. I might have to sell the shop soon...;Pierre
+Hey, that's a great idea... It'll be nice to get out from behind the counter. I'll meet you there!;Pierre
+Welcome! If you're looking for seeds, you've come to the right place!;Pierre
+Hi, <PLAYER>. Need any seeds or fruit tree saplings?;Pierre
+Hi, <PLAYER>. How's Abigail doing? The shop hasn't been the same without her...;Pierre
+Sometimes I get new items in stock, so make sure to stop by every so often. It's a lot of work to run a shop.;Pierre
+You know, I used to be a pretty good boxer back in the day... my right hook was the stuff of legend!;Pierre
+It's my day off today. Hmmm... what to do... If you find any wild produce, I'll take it off your hands for a reasonable price.;Pierre
+sigh ... No matter how well the business does, it's never enough to satisfy my desire for success. What's wrong with me? Maybe I'm just bored with domestic life. When I was younger I always had something interesting to do... now I find myself just wasting time whenever I'm not at work.;Pierre
+I'm happy to buy any produce off you. I'll give you a fair price, of course!;Pierre
+I really do appreciate your business, Mr. <PLAYER>. I've been having a harder and harder time turning a decent profit.;Pierre
+Does Abigail look anything like me? Don't tell my wife, but sometimes I wonder if I'm really the father. Don't tell my wife, but I hate to cook dinner.;Pierre
+Business has been slow since Joja moved into town. It's hard to compete with their selection. Hi there. Is your farm doing well? Maybe a few of my seeds will spruce things up.;Pierre
+If you find any wild produce, I'll take it off your hands for a reasonable price.;Pierre
+I'll tell you a secret. Sometimes, when the store is empty, I'll practice my opera singing behind the counter. Don't tell anyone.;Pierre
+sigh ... I've got those behind-the-counter blues... I wish I could go for a walk, but I can't leave the store unattended. I've heard the flowers are in bloom and the air smells great...;Pierre
+Some day I'm going to open a new shop in the city and make a killing. Unfortunately, business is so slow I can hardly save any money... Business has been pretty good since JojaMart closed down.;Pierre
+Hey, <PLAYER>. You're my number one customer!;Pierre
+I've got summer seeds available! You'd better stock up.;Pierre
+I'm so glad to have a day off from work. Now that Joja is gone, I feel a responsibility to keep the shop open seven days a week. People need groceries!;Pierre
+Running the shop is one of the most important things in my life, but sometimes you just need a day off... Now that Joja is gone, I feel a responsibility to keep the shop open seven days a week. People need groceries!;Pierre
+Everyone's enjoying the weather instead of shopping... sigh;Pierre
+Oh, hi. Are you satisfied with the quality of my summer seeds?;Pierre
+Summer is a good time to eat a refreshing salad, don't you think?;Pierre
+You can't grow any crops during winter... so I have no seeds to offer you. It's really going to cut into our profits.;Pierre
+Finally... a day of relaxation. Now that Joja is gone, I feel a responsibility to keep the shop open seven days a week. People need groceries!;Pierre
+I'm going to surprise Caroline tonight with dinner and a bottle of Stardew Red. Don't tell anyone! Hi. In the mood for a little shopping?;Pierre
+I mainly deal in edibles, so if you've got any ores or wood you'd better try selling them somewhere else.;Pierre
+Is it just me or is there a cold draft in here?;Pierre
+If I didn't have this shop to run I would be outside right now. It would be fun to throw a snowball.;Pierre
+Fall seeds are here! Crops don't grow in winter so this is your last shot until spring. Better go all out, huh?;Pierre
+It's starting to get cold outside... That means more people come into the shop to warm up... Hehe.;Pierre
+My dream is to keep opening new stores until I become extremely wealthy. Hehehe. Hi there. Is your farm still doing well? Maybe a few of my seeds will spruce things up.;Pierre
+I'm happy to buy whatever you produce on the farm. Yes, I do resell it for a profit... but that's the benefit of being an established shop owner. Don't get any ideas...;Pierre
+If I didn't have this shop to run I would be outside right now... The leaves look great and the mushrooms are sprouting.;Pierre
+Thanks, I appreciate that. Just forget you ever saw this, okay? ...And I'll just forget that you were snooping around in my bedroom!;Pierre
+You would really do that to me? You're terrible. You shouldn't have been snooping around in my bedroom in the first place! Nosy neighbors are the worst.;Pierre
+It feels great to be on top! Hah! Nice effort, though.;Pierre
+Have you met everyone in town yet? That sounds exhausting.;Robin
+Wow, look at these structures! I can tell they were made by a master builder. I'm really impressed!;Robin
+Wait a sec... you're giving me this? What a special treat! Thank you.;Robin
+What's this... 'Woody's Secret'? ... Um, what kind of book is this, <PLAYER>? Oh, it's about woodcutting! Hahaha... Oh yeah, I'm going to love this. Thanks!;Robin
+Ah, 'Woodcutter's Weekly'... I was hoping to get my hands on the newest issue! They've got all the latest techniques in here! From one woodcutter to another... Thank you.;Robin
+That man is insane! Why does he have to go out there? We have no idea if this rain is dangerous!;Robin
+I hope Demetrius is happy with the 'samples' he collected. Seems like his little science project is more important to him than his own wife. Hmmph.;Robin
+There's all kinds of good wood out there today. Better get chopping, huh?;Robin
+Sebastian's back in the basement... what happened between you two?;Robin
+Maru was crying through the night... We're all devastated...;Robin
+Oh, don't worry... these are tears of joy. I know Sebastian will be very happy on <FARM> Farm!;Robin
+We'll all miss having Maru around the house, but I'm sure we'll still see her often!;Robin
+It was fun working on your house. That might be some of the finest construction I've ever done! I hope you're enjoying the new kitchen.;Robin
+What are you going to do with all that new space in your house? Or perhaps I should call it... your 'mansion'?;Robin
+So, are you getting into the aged wine business? I've heard it's very lucrative... if you have the patience.;Robin
+That's probably the sturdiest coop I've ever built. Hope you get a lot of use out of it!;Robin
+So, what are your plans with the new shed? The possibilities are endless!;Robin
+How's that new coop treatin' ya? I'll be excited to meet your new animals!;Robin
+Are you satisfied with the new barn? I'm sure you'll raise some very happy livestock in there!;Robin
+phew ... Building that slime hutch was a lot of work... I'll need a day at the spa to recuperate!;Robin
+So, have you been riding your new horse around town? I imagine you're pretty sore, but you'll get used to it soon.;Robin
+I mentioned to Willy that you got a new fish pond. He seemed really happy about it. Apparently, he raised crabs when he was a young man. Called it a 'time-honored tradition'.;Robin
+Hey <PLAYER>. You know, I was just thinking about how far the town has come since you first moved here. You should really be proud of all you've done!;Robin
+Veggie salt? Sounds good, but I don't need it.;Robin
+Hmm... when I think of flowers, I think of Evelyn. Maybe show this to her?;Robin
+Our little plan worked out well, don't you think? Pam and Penny seem really happy.;Robin
+Ooh... Two tickets for tonight's show, huh? Sounds good! I'll meet you there.;Robin
+Hey! I heard some weird noises last night, and woke up this morning to find the quarry bridge completely repaired! It's a miracle of woodworking!;Robin
+I knew it from the moment you stepped off that bus, <PLAYER>... you were destined to be a town hero!;Robin
+Hey! I heard that Lewis showed you around the Community Center... It'd be great if you could fix that place up. It used to be a really nice building!;Robin
+Have I told you that I built our house from the ground up? It's definitely been the highlight of my career so far.;Robin
+Hey, if you need any materials or blueprints, my shop is the place you're looking for! Plus, your business supports the local economy.;Robin
+Hey there, <PLAYER>. I was just daydreaming about some new carpentry projects. Are you doing OK at your place? How's the woodwork in your cottage holding up?;Robin
+You're always welcome to visit us, even if you aren't shopping, you know. It can get pretty lonely up here in the mountains.;Robin
+You've met my son Sebastian, right? He lives downstairs. He's a little shy, but I'm sure he'll warm up to you if you're nice to him.;Robin
+I miss my Sebby... he was always a little misunderstood, but I believed in him.;Robin
+I found an ashtray in Sebastian's room, and it smelled really weird. Should I be worried about this? Sometimes I worry about Sebastian... he doesn't have many friends and doesn't really seem to care about his future very much... I would talk to him about it but he never opens up to me.;Robin
+Sebastian told me he's trying to quit smoking! I'm really proud of him.;Robin
+Our house is in such a beautiful area, don't you think? I love the fresh air of the mountains.;Robin
+I hope Demetrius doesn't blow the house up with those science experiments of his. I'm not even sure what he's working on. I think it has something to do with plants.;Robin
+My husband almost set the house on fire last night with his science experiment. One of his beakers exploded and sent a fireball into the rafters! Thank Yoba I used fire-resistant lacquer when I built the place. Sorry if it smells weird in here, <PLAYER>. It's my husband's bizarre science project...;Robin
+Hi <PLAYER>! It's good to see you. You're a regular member of the community now, don't you think?;Robin
+Great, then this blueprint should be a piece of cake for you!;Robin
+Well, then I have the perfect blueprint to get you started!;Robin
+Wood is a wonderful substance... it's versatile, cheap, strong, and each piece has its own unique character!;Robin
+If you need any buildings on your farm upgraded, just ask me! You'll need to provide enough lumber and stone for the project. And it costs money, too. But I'm sure you'll be pleased with the results!;Robin
+We're pretty insulated from the rest of the world here in Stardew Valley. It has its pros and cons.;Robin
+Maru likes gems. She uses them in her inventions. So how are you and Maru getting along?;Robin
+You're always welcome to visit us, even if you aren't in the market for a building upgrade! It can get pretty lonely up here in the mountains.;Robin
+Have you met the wild man that lives behind our house? I guess I don't really mind, as long as he doesn't bother us.;Robin
+Hey, it's a perfect time of year to gather wood. Maybe you can collect enough to upgrade your house!;Robin
+My parents were bewildered when I told them I wanted to be a carpenter. They were pretty old-fashioned.;Robin
+I think the farm life is really healthy for Sebby.;Robin
+Our house is in such a beautiful area, don't you think? Everything looks still after a fresh snow.;Robin
+Did you watch the game last night? Or wait, do you even have a TV set...?;Sam
+Ugh, I stepped in something gross earlier... and I just bought these shoes.;Sam
+Aw yeah... Pizza time!;Sam
+Hey, hey, hey! You remembered! Now that's what I'm talking about!;Sam
+Hey, hey, hey! You rememb-- ... sniff ... sniff ... what's that gnarly smell...? ...Oh.;Sam
+Oh... with me? Hehe... Yeah, sure. Sounds fun!;Sam
+Whaa! Is it acid? Will it burn through the roof and sizzle us alive? I gotta keep my little brother safe!;Sam
+Seems like there's no damage to anything... That was a strange one. How are you holding up?;Sam
+You notice how warm this rain is? It feels alive...;Sam
+Um... So, can I get a kiss now?;Sam
+I couldn't sleep last night. I was imagining all the things we're gonna do together.;Sam
+Whoaa... Hold on... this is for me? I never expected this! Thank you!;Sam
+I heard you and Sebastian are going out... he better still be coming to band practice every week!;Sam
+Wow, I can't believe you and Sebastian got hitched. Times really are changing... Eh, it's alright, I'm sure he'll still have time to hang out.;Sam
+Hey, what's the big idea? You took my drummer! Haha, just kidding... Congrats!;Sam
+I better lay low for a while. I wouldn't want to get on Mayor Lewis' bad side!;Sam
+Heheh... So you've been to the sewer, huh? I've snuck down there a few times myself. It's pretty creepy. How could I tell? Hmm... let's just say... 'the nose knows'.;Sam
+Wouldn't it be cool if the town had a skate park, though? Now, how to convince Mayor Lewis...;Sam
+Even with all the practice we did, I made a bunch of mistakes on stage... I hope no one noticed. It's kind of nerve-wracking to be in front of everyone!;Sam
+If... I mean when... my Dad gets back, I wonder if he'll be the same as I remember him?;Sam
+I bet there's some cool stuff deeper on the island. Willy said we have to stay here, though... some kind of liability issues or something.;Sam
+I'm feeling pretty good from all the sunlight.;Sam
+Check out these gnarly waves... Guess what? I'm gonna save up for a surfboard.;Sam
+Maybe I should learn to play the ukulele. Abby on the congas and Sebastian on the steel drums... 'The Pelicans of Paradise'... What's the matter, you look worried.;Sam
+I'm just getting juiced up, then it's back to the beach.;Sam
+Aw, I gotta put on that dorky suit again? I thought now that we're married... Well, alright.;Sam
+Alright, I'm done with the cold shoulder act. Let's just forget about what happened.;Sam
+Wait... what?! I thought we were doing okay... I guess I was way off... alright, then. Yeah, sure... we can still be friends.;Sam
+I'm sorry, <PLAYER>. I don't think I can still be your friend.;Sam
+Hey, how's it going? Last night I practiced guitar for 4 hours straight. My fingers hurt like crazy. Bye, I've got something to do...;Sam
+I'll tell you a secret... my name is short for something else. Most people don't know that so keep it on mute.;Sam
+Oh, hi! It's good to see you again. Take it easy.;Sam
+Hi, <PLAYER>. Do you want to hang out or something? Sleeping feels nice but sometimes I wish I could skip it... it's just such a time-waster.;Sam
+I'm trying to come up with a new song for my band, but I'm blanking... Hey.. What do you think my new song should be about? ;Sam
+But I already know what you think about it. Thanks to your help I'm already done with the lyrics!;Sam
+Hey, man... what the heck is your problem? Sorry I asked.;Sam
+Hey, you know what? That sounds perfect. Thanks!;Sam
+I've been friends with Sebastian for a while. He's a pretty cool guy. If you two weren't around I wouldn't really have any friends in town.;Sam
+Hey buddy. You look like you're full of energy today!;Sam
+I think my Dad might be coming back in a year or so. I guess you might be able to meet him, then. If you're still living here. You wouldn't abandon the farm after putting so much effort into it!;Sam
+...I have mixed feelings about my father being back. He seems strange. But I guess I haven't seen him in 3 years... so that could explain it.;Sam
+The clouds look great this time of year, don't they? Like scoops of vanilla ice cream floating on blue raspberry sauce... Or maybe I'm just hungry.;Sam
+My Dad is a soldier, fighting against the Gotoro Empire. That's why he's not here... He'll come back someday. ...I've heard some terrible things about the Gotoro Empire...;Sam
+Well, my Dad is back. Have you met him? I'm just glad he's okay.;Sam
+Did I ever tell you we used to live in the city? There's a lot more stuff going on, but it's also kinda noisy and hectic... It's hard to say which is better.;Sam
+Dad used to work as a garbage man in the city. You don't even want to know some of the things he found in people's trash.;Sam
+The older I get the more attached I feel to the valley. Although I could be biased... for a certain reason. So, how are you doing today?;Sam
+Oh! I just remembered I'm supposed to call my Grandma. Okay, I'm going to put this rubber band on my wrist so I don't forget. I have to make little reminders for myself or else I'll totally forget to do things.;Sam
+Hmm. I just remembered that I was meaning to do something... But I forgot. This happens to me all the time.;Sam
+Sometimes I wonder if my mother is actually happy with her life... I asked her once and got the feeling she was hiding some inner pain.;Sam
+Sometimes those spring petals that float around make my nose drip. I like summer best, myself.;Sam
+Hey, you know what? That's exactly the kind of style I've been thinking about... for the band.;Sam
+My little brother gets so excited for festivals. He's been talking about tomorrow's egg hunt all day.;Sam
+Why couldn't the flower dance be today so I could get work off?;Sam
+One time, Gus fell in the river during the Ice Festival. We had to build a huge fire in the saloon to dry him out. We were all worried we'd never taste his famous spaghetti ever again.;Sam
+Tonight, after everyone is asleep, that weird old man who lives in the woods will set up the haunted maze. He always makes sure no one is around to see him working.;Sam
+I don't have to work tomorrow! I've never been so happy.;Sam
+The Feast of the Winter Star is supposed to be a time to show gratitude for all the good things in life. Me? I'm mostly interested in the food. Hehe.;Sam
+Have you been to the desert yet? You can find some interesting things there. Dad took me a couple times when I was younger.;Sam
+There's a little less pollen in the summer... my nose is really happy.;Sam
+Hi, <PLAYER>. Do you want to hang out or something? I'm so energized during summer I hardly have to sleep!;Sam
+I saw a spider on my ceiling last night. They sure like the warmth. I hope he didn't crawl into my mouth while I was asleep...;Sam
+Dad was telling me about our national enemy, the Gotoro Empire... They sound like a ruthless bunch.;Sam
+Sometimes I can hear my Dad pacing down the hallway at night. ...But he won't tell us what's troubling him.;Sam
+You know... I used to live in the city and I have to admit, this place is much nicer in the summer. The city is so hot with all that concrete.;Sam
+When you see hot air balloons drifting over the sea, you'll know that summer's in full swing.;Sam
+I like taking walks in the forest this time of year. The temperature is so comfortable. ...;Sam
+My shoes always seem to wear out this time of year. Look, my toe's poking through! That's what happens when you skateboard too much.;Sam
+I ordered a new pair of shoes online. They're the same style so I know they'll fit right.;Sam
+I ordered some shoes online more than a week ago! Man, they're really taking a long time. I guess we must really live in the middle of nowhere.;Sam
+Ah, my feet feel great in these new shoes. They're a little crisp, but they'll break in. Especially after I run through the haunted maze this weekend.;Sam
+Hi, <PLAYER>. Do you want to hang out or something? Just stop by whenever you want.;Sam
+Hey... something smells good, like pizza. Do you smell that? Maybe Mom's cooking something.;Sam
+I think my Dad might be coming back next year. I guess that's kind of soon. Weird.;Sam
+I guess I should start thinking about moving out soon... ...But I would miss Mom's fish casserole, so I dunno... I hate making decisions.;Sam
+Did you ever notice all the spider webs that appear during the fall? If you get up early you'll see them covered in dew... like little stars.;Sam
+I used to live in Zuzu City, you know. The biggest difference is the noise. It's never quiet in the city. There's way more people around you, always. And it's always bright with city lights, even at night. I guess which setting you like depends on your personality.;Sam
+I think I've grown to like the country more than the city. Although I could be biased... for a certain reason. So, how are you doing today?;Sam
+I'm lucky to have the biggest room in the house to myself. I guess there's advantages to being the oldest child.;Sam
+Hmm. I think I was meaning to do something... But I forgot. This happens to me all the time.;Sam
+The first week of fall is already over. Man, that went by fast. Whatever. I'm just happy that you came to visit me! So do you have anything new to tell me?;Sam
+It's hard to have fun on Sunday... I keep thinking about having to go to work tomorrow.;Sam
+Hi <PLAYER>. Hmm.. I wonder if I could grow a cactus in my room?;Sam
+Everyone's indoors during the winter, so I can play guitar as loud as I want.;Sam
+Hi, <PLAYER>. Do you want to hang out or something? Well, it's a good time of year to catch up on all the sleep I've missed.;Sam
+Joja I hate working at JojaMart, but how else am I going to make money in this town? I'm glad I work at the museum now. It's way more interesting than that depressing JojaMart ever was.;Sam
+I think my Dad is coming back this spring. It's been 3 years since I've seen him. This is going to be strange.;Sam
+Dad's been back for a while now. I guess I'm finally getting used to it.;Sam
+My Dad is in the military, fighting against the Gotoro Empire. That's why he's not here. He's coming back soon, though.;Sam
+One thing that's nice about winter... all the mosquitos are gone.;Sam
+The worst thing about the snow is getting your feet soaked. You're lucky in those thick boots.;Sam
+I just remembered I need to change the strings on my guitar. Okay, I'm going to write a note on my wrist so I remember to do it tonight. I have to make little reminders for myself or else I'll totally forget to do things.;Sam
+It would be fun if we could go sledding, wouldn't it? I don't think there are any good hills nearby. Plus, I don't have a sled. So, how have you been doing lately?;Sam
+Try not to make a mess in the house or my Mom will flip.;Sam
+A Stardew Valley rose? That's so lovely! Thank you! Whenever I miss the valley, I'll smell this rose. Hehe. Here, take this. It's been on the shelf for 10 years and no one has bought it. You can have it!;Sandy
+Oh, wow! Nothing says 'spring' like a Stardew Valley daffodil!;Sandy
+Wow, that's so rare... I can't believe you're giving this to me! Thanks, honey!;Sandy
+Ahh... even in winter, Stardew Valley has beautiful flowers.;Sandy
+I heard there's a rare weather event in the valley. I wish it rained here sometimes...;Sandy
+How'd the big day go? Emily came up a while ago and told me all about your plans. I'll have to come visit you two on the farm sometime!;Sandy
+sniff ... Mmm... sweet peas have the most wonderful smell.;Sandy
+Hello hello, sweetie! I'm so glad that the bus is back in service. It was starting to get lonely out here. ...I almost went out of business. Hey, but now you're here to buy a whole bunch of seeds... right, sweetie?;Sandy
+You can find coconuts and cactus fruits outside. Feel free to take as many as you can find. They're oooh-yummy, too. Heeee!;Sandy
+Why hello there! If you see my friend Emily in town, could you tell her I said 'Hi'? Thanks!;Sandy
+That shady-looking guy in the back? I'm not supposed to talk about it. Anyway... what's the weather like in the valley?;Sandy
+You wouldn't have much luck farming out here, honey. Well, unless you were a cactus farmer.;Sandy
+Hi! Welcome to my Oasis, the number one destination in Calico Desert! whisper Did I sound professional? Make sure to check out our seasonal seed selection. You look really thirsty. Here, have a taste of this ice cold milk. Tasty, huh?;Sandy
+It's always sunny here, even in winter! Actually, it's a little hot for my tastes... my delicate skin burns quickly in the sun.;Sandy
+That shady-looking guy in the back? I'm really not supposed to talk about it. Anyway... what's the weather like in the valley?;Sandy
+Hey... I think I can trust you. You can keep a secret, right? whisper That guy in the back... it's some kind of exclusive club. I've never been inside myself, but they pay me a generous monthly rent so I don't ask any questions. SOOOOoo! How's that Stardew Valley weather right now?;Sandy
+My name's not real, you know. It's just good for business. But everyone calls me like this anyway, so I've just gotten used to it.;Sandy
+Isn't it expensive, coming to the desert so often like you do? You must really like to see me, huh? Hmmm... sigh ... I do miss the beautiful flowers of Stardew Valley...;Sandy
+Hi there, honey. Yep, it's just me, alone in this big old shop. Hmmm... Well, except for that creepy guy in the back. But he never moves a muscle, so I often forget about him.;Sandy
+Hey, it's <PLAYER>! I'm glad to see you, kid. I was starting to think you'd never come back! If I didn't have this shop to run I'd come back to the valley with you for a day or two... just you and me on the old farm. You'll just have to visit me more often so I don't get depressed.;Sandy
+Hi! Stay as long as you like. The air conditioning sure feels nice, doesn't it?;Sandy
+Oh. You just moved in, right? Cool. Out of all the places you could live, you chose Pelican Town?;Sebastian
+Oh, really? I wasn't expecting this. It looks great, though. Thank you.;Sebastian
+Don't make fun of me... but I've never been in a relationship like this before.;Sebastian
+Hey... you look great. I've been thinking about you all day...;Sebastian
+Hey... It's been a year since we started dating, remember? It's been the best year of my life...;Sebastian
+Oh yeah? ... Alright, I will. Heh... maybe this festival isn't so bad.;Sebastian
+Demetrius has been all wound up because of the wedding, it's been pretty tense around the house.;Sebastian
+My mom's freaking out, but I'm actually kind of enjoying this. I bet the frogs are going wild out there! I'm gonna stay here to try and keep her calm, though.;Sebastian
+I was hoping this weather would come back. It makes me appreciate the summer a little more.;Sebastian
+The show went well. I was kind of nervous beforehand, but once we went up on stage I just kinda got into the zone.;Sebastian
+Is that obsidian? This is exactly what I've been looking for... I'm gonna see if I can shape it into a dagger.;Sebastian
+Hey, thank you. I'll drink this when the sky gets dark.;Sebastian
+What are you trying to say... I need to work out? Well, I'm more of a 'nimble' guy... Maybe I'll read the first half.;Sebastian
+I really love this. How did you know? Mmm... once you get used to eating raw fish, it becomes really addictive.;Sebastian
+Whoa... This egg speaks to me... I'll put it on my desk. Thanks! Hmm... I wonder, if I put this under my pillow, will it hatch? Hehe... only one way to find out.;Sebastian
+You brought me soup? That's really nice of you. This one brings back a lot of memories...;Sebastian
+Hey, that's a frog egg! I'll see if I can get it to hatch. Thanks!;Sebastian
+I'm glad the path to the railroad tracks is open again. I like going up there when I feel like being alone. ...Which is often.;Sebastian
+I liked the boat ride. At one point you couldn't see any land at all..;Sebastian
+The beach is nice, but I don't love the crowds.;Sebastian
+You think Willy would let me lug a jet ski out here? ...Nah.;Sebastian
+...Now, if I were a tropical frog, where would I hide...;Sebastian
+I bet this place looks pretty cool in the rain.;Sebastian
+It's good to get away from the computer and see the real world sometimes. I forget how nice it can be.;Sebastian
+I usually wear black, but in this heat...;Sebastian
+No... There's someone else I want to dance with.;Sebastian
+Why do you keep bothering me? I'm trying to forget about you.;Sebastian
+If I just disappeared would it really matter?;Sebastian
+I was thinking... people are like stones skipping over the water. Eventually we're going to sink.;Sebastian
+Sam's really outgoing, so when we hang out he does most of the talking. That's fine with me! Imagine if Sam was as reserved as I am... we'd hang out and never say a word. Some people are talkers, others are listeners. I'm more of a listener.;Sebastian
+What am I going to do today? Probably nothing.;Sebastian
+Sam is probably my only friend in this town. Well, Abby is nice too, but... Umm.. never mind.;Sebastian
+You and Sam are probably my only friends in this town.;Sebastian
+I dunno where I'd be without you... probably in a ditch somewhere.;Sebastian
+I snuck into the caves last night and got a nasty cut from a rock crab. Don't tell anyone, okay? Hi. ...want something?;Sebastian
+I come here for the peace and quiet.;Sebastian
+Why does everyone like Maru so much? Sure, she's smart and friendly, but don't they realize it's all just an attention-grabbing scam? Sorry...;Sebastian
+Could you picture someone like me living on a farm? It seems ridiculous.;Sebastian
+Hey... How are things going on the farm? Same old routine, huh? Yeah, I know a thing or two about that... Could you picture me living on a farm? It seems ridiculous, but I have been thinking about it lately.;Sebastian
+sigh ... If gas wasn't so expensive I'd ride my motorcycle to the city today. So what do you do when you aren't working?;Sebastian
+Hey, if you're bored feel free to hang out in my room... You probably prefer the small-town life, huh?;Sebastian
+Oh yeah? Well, did you read the new 'Cave Saga X'? I won't spoil it for you, but oh man...;Sebastian
+Oh yea? Hmm... Not really my thing, but everyone's different I guess.;Sebastian
+What? I didn't hear you... I'm busy thinking about something. What do you want?;Sebastian
+You know, I should be doing something productive right now. I just lose focus too fast... Maybe I should drink more coffee?;Sebastian
+Having a good weekend? ...nice.;Sebastian
+I usually stay inside, but I do go to the beach now and then. Pretty much only when it's raining, though. For some reason, staring off into the bleak horizon makes me feel... I dunno. Like it's worthwhile to keep pushing on, I guess.;Sebastian
+What's new? I'll be at the saloon tonight destroying Sam at pool again.;Sebastian
+Good choice. That works well with my role.;Sebastian
+Not so fast! I want to know what it is.;Sebastian
+Really? It was making a lot of noise!;Sebastian
+It's too dangerous without a sword. But I like your sense of adventure.;Sebastian
+Don't worry, I'll make sure to show you the ropes before I let you do anything dangerous.;Sebastian
+Hmm... Not exactly how I'd put it but, yeah.;Sebastian
+Yea... That's exactly how I feel, too.;Sebastian
+Oh... Well maybe this wasn't the right place to bring you, then. I thought you'd see the beauty in it.;Sebastian
+Ugh... You sound like my Mom. ... sigh You're right, though. I'm addicted to these things, it's not good.;Sebastian
+Everyone is so happy when it's sunny out... Different people like different things, I guess. Most people like the sun, huh? I guess you do need a minimum amount of vitamin D...;Sebastian
+I'm looking forward to the cold, damp season. I feel more at home.;Sebastian
+I'm looking forward to the cold, damp season. Channeling my inner toad, I guess.;Sebastian
+What's the point of going outside? I've already seen it all.;Sebastian
+If you're depressed, everything seems stupid. But if you're feeling positive, then even the most mundane things will make you happy. How do you stay positive long term? I have no idea. ...Maybe I'll figure it out some day.;Sebastian
+Why am I grinning? I guess I've been unusually happy lately...;Sebastian
+I definitely don't want to go swimming, if that's what you're going to ask. Oh, you just stopped to say hi? Sorry.;Sebastian
+If you ever wanted to go swimming... I guess I'd do it.;Sebastian
+The temperature only starts to get comfortable at night. It always seems to stay a little cooler by the lake.;Sebastian
+Hey, I'm glad you stopped by. How have you been coping with this heat?;Sebastian
+I hope we're not having mushroom casserole tonight... Oh no, is it cave carrot tacos night? ... groan;Sebastian
+I'm not too big on festivals. Lots of commotion, too much social interaction... More stress than fun, if you ask me.;Sebastian
+Ugh... Tomorrow is the worst festival of the year by a mile. Just wait and see how stupid I look during the flower dance.;Sebastian
+Are you looking forward to the Spirit's Eve festival? It's probably my favorite.;Sebastian
+The Spirit's Eve festival is tomorrow. I really hope they have those skeletons on display again this year. I almost feel like they're my friends at this point.;Sebastian
+Hehe... It might be funny to put something disgusting in the potluck soup tomorrow. Just don't poison anyone. Why ruin the potluck? Hmm... I guess some people feel liberated when the rigid structures of society break down a little. Maybe I'm weird.;Sebastian
+Tomorrow we'll all be gathering at the docks to see the migration of the moonlight jellies. I think it starts at ten o'clock.;Sebastian
+It's the eve of the winter star today. Better make sure you've got an empty stomach for the feast tomorrow.;Sebastian
+I usually only go outside after dark. Does that sound weird to you?;Sebastian
+Last night, I saw a group of bats flying around near the lake. I guess maybe they are feeding on the mosquitoes this time of year?;Sebastian
+So... it's the big harvest season, isn't it?;Sebastian
+You're probably making a lot of money on your farm, huh? I guess I should get a job soon...;Sebastian
+Pumpkin spice this... pumpkin spice that... man, I do get sick of these seasonal fads.;Sebastian
+Who does Demetrius think he is, telling me what to do? He's not even my real father.;Sebastian
+I've always felt like an underdog. But the challenge will just make me stronger.;Sebastian
+yawn ... what time is it? I think I slept too much last night.;Sebastian
+I'm glad you showed up. I was starting to get bored. Now I feel a lot more energetic, believe it or not.;Sebastian
+My room might be a little less depressing if I had a window...;Sebastian
+yawn ... I was up until 3 reading this new book... Do you read, <PLAYER>?;Sebastian
+You probably wouldn't like it. I think you'd like it.;Sebastian
+Hey, me too. There's nothing like a good fiction to push the boundaries of the imagination.;Sebastian
+I've been reading a book about a farmer. Reminds me a lot of you, actually. Sorry if that creeps you out.;Sebastian
+I was this close to moping in bed the entire day. Kinda wish I had, now.;Sebastian
+The sun is coming out less and less these days... wonderful.;Sebastian
+I'm enjoying this gloomy weather... although I have to say, it's a little cold.;Sebastian
+Nothing surprises me anymore... nothing makes me laugh. Yeah, I know... I'm being a little dramatic.;Sebastian
+Do you ever think back on your old self with shame and embarrassment? I guess it's just part of growing up.;Sebastian
+Hmm... I don't really have anything to say. I still don't know you very well.;Sebastian
+Hey... It's hard to think of new things to talk about, sometimes. Even after you know someone. ...Sorry.;Sebastian
+I was so close to screaming at Mom for throwing away my old comic collection... But something stopped me. Hmm... with age comes wisdom.;Sebastian
+Hey, don't let me stop you from getting your work done. If you aren't busy I don't mind if you stick around.;Sebastian
+I wonder what's for dinner tonight... grumble;Sebastian
+I should learn how to cook. That would be useful.;Sebastian
+It's fun to live so close to the train tracks. It's a reminder that there's a big world out there.;Sebastian
+When I was younger I'd run out to the railroad whenever I heard a train coming. Whatever happened to those days...? sigh;Sebastian
+The frogs aren't very happy in winter. Poor little guys. I wonder what the frogs do in winter? Do they just become frozen and then thaw out in spring?;Sebastian
+I like winter but I do feel bad for all the cold animals out there. I guess they're used to it, though... like me.;Sebastian
+I built a snowgoon but Demetrius made me get rid of it. What's his problem?;Sebastian
+I built a snowgoon but Demetrius made me get rid of it... yet Maru's cute little snowman goes unharmed! Ridiculous.;Sebastian
+I don't know you. Why are you talking to me?;Shane
+Oh, is it my birthday? I was hoping I'd forget. This is actually really nice, though. Thank you.;Shane
+Oh, is it my birthday? That's right... I almost forgot. Thanks.;Shane
+I don't have to go into work today, so I'm not complaining...;Shane
+Oh, there's some kind of weird weather today? I didn't notice.;Shane
+Heh... thanks. Tastes just like beer.;Shane
+Um... sure. I'd better stop eating...;Shane
+Oh, you made it? squish ... Um... smack ... I think I'll stick to frozen pizza. But thanks.;Shane
+Have you tried the new Joja Bubblerito? I ate two in one sitting yesterday, that was a mistake...;Shane
+I... wasn't really expecting this, after all I've put you through...;Shane
+I think I might be the luckiest guy in the world...;Shane
+Sorry you had to see me like that, <PLAYER>...;Shane
+I'm trying to cut back on the drinks... it's not going to be easy.;Shane
+Buh... hope the bar's open.;Shane
+Welp... got my year's supply of Vitamin D. Now I can play video games with the blinds shut the rest of the year.;Shane
+That's right... I'm not wearing any shoes or socks. Some call it a public health risk. The crabs seem to enjoy it, though.;Shane
+Hoo... Feels good to rest these hams. If anyone needs me, I'll just be sitting here, marinating.;Shane
+Phew... feels good to have a cold drink... Or maybe a few...;Shane
+Okay, this sand is piping hot. I regret not wearing shoes.;Shane
+Dance? Hoo-boy. Okay... let me chug a few more of these first...;Shane
+I heard you built Pam a house... That's really generous... You know, I could use a hundred thousand gold myself... ahem ...ah ...Just kidding. ...Heh.;Shane
+I heard someone bought Pam a house... sigh ... What did Pam ever do to earn that kind of generosity??;Shane
+... ...Leave me alone.;Shane
+gulp Oh... Yeah, okay... I guess I should've seen this coming. I'll... uh... guess I'll see you at the Saloon.;Shane
+I hardly know you. Why are you talking to me? What's the point of talking to me? I don't want any friends.;Shane
+Hey. Sorry if I came off as rude when we first met. It takes me a while to warm up to strangers. It's a lot of effort to be friends with someone. It's easier to just be alone.;Shane
+I'll be honest, I still can't understand why you'd want to be friends with me. Life isn't fair, and I can't help but feel bitter about it.;Shane
+Do you ever feel like the whole world is scrutinizing you like some disgusting insect?;Shane
+What? What do you want? Go away.;Shane
+What do you want from me? Money? I'd give you a pot of gold to leave me alone!;Shane
+Every time I try something new it goes horribly wrong. You learn to just stay in a shell.;Shane
+Life isn't so bad... at least I have frozen pizza and eggs.;Shane
+We should go to a Tunnelers game some time. When I go by myself, I usually drink way too much beer.;Shane
+I need to keep it together for Jas' sake. I don't want to be a bad role model for her. I'm kind of like a father figure to Jas. I just hope I haven't let her down too much.;Shane
+No, I don't have time to chat with you.;Shane
+I guess I've grown attached to Marnie and Jas. We're a ragtag bunch but it kind of feels like a weird family. I never really had much of a family as a kid.;Shane
+sigh ...Every day is the same. Stocking those horrible shelves, going to the saloon, tossing and turning all night. ;Shane
+What do you want? Leave me alone.;Shane
+You're really persistent. I guess I'm just surprised that anyone would be interested in talking to me.;Shane
+Buh... Where's the Joja Cola? Why try? Life's just a roll of the dice. When they rolled mine, the dice fell off the table.;Shane
+You like talking to me? I guess I believe you... maybe you're as weird as I am.;Shane
+What's the point of life if all you do is work? I haven't been able to pay my rent, but Marnie's letting me stay anyway. I'm lucky.;Shane
+I'll definitely be going to the saloon tonight. ;Shane
+Buh... Can you imagine how disgusting my liver must look? My body feels twenty years older than it should...;Shane
+My body hurts a lot less since I cut back on the booze. But it still feels like I'm always teetering on the edge...;Shane
+Why are you bothering me? I want to be alone.;Shane
+I'm surprised that you're still trying to make friends with me. Haven't I been rude enough to you yet? sigh ...;Shane
+I'm renting my room from Marnie at a really good price. It's small but I can't complain. If I could reset my life maybe I'd start a chicken farm. Only free-range eggs of course.;Shane
+I've really grown attached to Marnie's chickens. A chicken will never let you down. Humans, on the other hand...;Shane
+Your farm seems like a great place to raise chickens.;Shane
+I'm busy, can't you tell? ...I don't feel like talking to anyone.;Shane
+You again? How many times do I have to tell you to leave me alone?;Shane
+Hmm... it's <TIME>. Should I throw a frozen pizza in the microwave, or should I wait? If the Tunnelers win, I'll knock back a cold one to celebrate. If they lose? I'll have a cold one to drown my sorrows.;Shane
+...I think I might be gaining weight.;Shane
+If I watch more TV maybe I can forget about life. That would be relaxing.;Shane
+Hi, <PLAYER>. I've actually been having a pretty good day so far.;Shane
+Okay... that'll require a great script. Maybe I'll ask Elliott for advice next time I see him at the saloon.;Shane
+Sounds great. I'll have to create a bunch of practical effects to make it look cool.;Shane
+I didn't know you were such a miser...;Shane
+Wow, it was that serious, huh? I can hardly remember...;Shane
+Oh yeah? I guess that makes sense. Me? I get bored with Pelican Town sometimes. But... I like that you're different. We balance each other out.;Shane
+Oh really? I'm surprised... Didn't you move to Stardew Valley to escape the noise of the city? I mean... Don't get me wrong, I totally understand. My life in Pelican Town is pretty bland, you know.;Shane
+Oh, a stranger! Momma says not to talk to strangers... But you seem okay.;Vincent
+Is that a birthday present? Can I open it now? Oh, wow! This is exactly what I wanted! Yayayaya!;Vincent
+A birthday present? Oh... this isn't very fun.;Vincent
+Wow! It's so sweet! Mmmmm....;Vincent
+Squish squish ... Ewww! Can I have it? Can I have it? Yaaay!;Vincent
+Mom! I wanna go outside and play!;Vincent
+sniff ... what is this? It smells awful... Mommmm!!!;Vincent
+That was fun! I wanna be a boat captain some day!;Vincent
+I wanna come back soon!;Vincent
+I wanna see a shark!;Vincent
+Don't take my Sammy away!;Vincent
+If you look close, you can see little bugs jumping out of the sand.;Vincent
+Oooh! Nice find! Mom wouldn't let me keep him, though...;Vincent
+I wanna be just like my big brother when I grow up! Okay! Bye!;Vincent
+I'm glad Dad's back... but he's different than I remember... Dad's not around much... I miss playing with him.;Vincent
+sigh ... Mom won't let me have any more gummies today.;Vincent
+Can you keep a secret, mister?;Vincent
+I wanna look for bugs, but Mom gets mad when I'm all dirty. It's a tough choice.;Vincent
+I'm hungry... where's Mom?;Vincent
+You're not as boring as most grown-ups!;Vincent
+Ew, it's boiled beet night again... Oh no... Mom's making lentil soup tonight.;Vincent
+Dad's coming back soon! I hope he brings me some toys.;Vincent
+Miss Penny makes me read a new book every week. At least she lets me pick them, sometimes!;Vincent
+Dad seems weird since he came back. I wonder what happened to him? He doesn't play with me anymore. I wonder if Dad will ever come home...;Vincent
+Hi there, mister!;Vincent
+Hi Uncle <PLAYER>!;Vincent
+Ahoy there! It's nice to see young folk movin' in to the valley. It's not very common these days.;Willy
+By the stem o' me pipe... a cup o' Stardrop Tea! And here I thought it was only the stuff of mermaid legend. Thank you kindly.;Willy
+Aye... A timeless classic. Brings a tear to me eye, with every read. A fisherman can never have too many copies! Thank ye.;Willy
+Thank ye! This'll be perfect winter reading. Ah, and it even smells of the sea...;Willy
+Ah, 'Bait And Bobber'... Every angler ought to have a subscription, but it costs a pretty penny. Taught me a lot of good fishin' tips over the years. Thanks!;Willy
+Ah, this stuff warms the ol' bones. Thank you.;Willy
+Aye... I hope these weird rains don't bother the fish... It could ruin me life!;Willy
+Well, the fish seem as healthy as ever. In fact, I swear they've been biting with extra vigor!;Willy
+If the fish like these rains, then I like 'em, too.;Willy
+Ah... That's the perfect bait for a catfish I've been tryin' to catch! Here, let me trade you this for it. I'm not sure what it is, but it washed up on the beach not too long ago. Somethin' tells me you could use it, lad.;Willy
+The horror! ...What kind of monster is this?;Willy
+chomp ... Yep, that's real gold alright. Heh! What a gift!;Willy
+Giving up the life, eh? Just don't forget to make time for fishing...;Willy
+Ah, you caught the Angler, did ya? A good 'first step' into the world o' rare fish, that one.;Willy
+So, you finally conquered the Crimsonfish! That fish has been taunting me for many years. He nearly nibbled through the wood under me house!;Willy
+By the stem o' me pipe... You actually caught 'the Legend'. I never thought I'd see the day... Well, lad, you've done it... You're the best fisherman this valley's seen in a long time...;Willy
+Ah, so you finally snagged the Glacierfish... That's one of the slipperiest fish I've ever encountered! How many tries did it take ya?;Willy
+sniff sniff ... Something tells me you've been fishin' in the sewers, lad.;Willy
+sniff ... Ah... That old familiar scent. You've been raisin' crabs, haven't ya? Yer a after me own heart, <PLAYER>. Hah! Here, take this. I found it on the beach... Been savin' it fer a special occasion. It's not every day I discover a fellow crab enthusiast!;Willy
+Ah... a movie. That's kind of ye, lad.;Willy
+Tomorrow is the annual Trout Derby... Aye, it's a freshwater enthusiast's dream! Anglers from all over the world will be here to catch our beautiful Rainbow Trout... I'd better see you out there, lad!;Willy
+It's the big day! I've got all me equipment tuned to catch Trout... Good luck out there!;Willy
+How's the Derby treatin' ya so far? I've found plenty o' Golden Tags, meself... but it's not over yet!;Willy
+Tomorrow, at the beach, is the start of a little gathering... called 'SquidFest'. Aye... tis' a difficult catch, the squid. But I trust that you've developed a knack for the sport, by now.;Willy
+Don't give up just yet! There's more squid in these waters...;Willy
+It warms me ol' winter heart to see all these anglers gathered here, all for the love o' fishing...;Willy
+Ahoy there, son.;Willy
+Have you had much luck in the local waters? You look like you could be a strong angler if you set your mind to it.;Willy
+A true angler has respect for the water... don't you forget that.;Willy
+The ol' Man was a fisherman, you know. I grew up sailing between the fern islands, living from catch to catch. The fishing life is tough, but there's something about it that keeps you goin'.;Willy
+Some fish come and go with the seasons. Others only come out at night or in the rain.;Willy
+I've heard a rumor that a special kind of fish can be found in the pools between waterfalls... If you can reach it with yer rod, that is...;Willy
+mumble mumble ...Eh? I would tell you about my thoughts, but it's a fisherman's secret.;Willy
+There's rumors of some very rare fish in these parts... but only an experienced angler could stand a chance against them. You'll need the finest bait you can get if you want a rare fish to bite.;Willy
+If you really want to get the fish biting, make sure you put some bait on your hook.;Willy
+I'm a simple man... when my pipe is full, I'm happy. When my pipe is empty... well, then I just fill 'er up again.;Willy
+If the local fishing scene got a bit more lively, I might expand the shop's stock.;Willy
+How's the fishing life going for ya? I try to fish as often as possible, but it's not easy when you've got a shop to run.;Willy
+If you purchase a high quality fishing rod, you'll be able to attach bait and tackle to the line. Bait causes fish to bite faster. Tackle can make your job a lot easier when reeling fish in. There's all kinds of tackle, with different perks. They're usually a little flimsy, so don't be surprised if your tackle breaks after a while.;Willy
+Hey, it's my skipper, <PLAYER>! I'd give you a discount on bait if I could afford it.;Willy
+I don't have high hopes of winning... the fish look nice but they don't exactly smell great...;Willy
+Wow, I got a low score. I guess Mayor Lewis isn't much of a fish man. Oh well...;Willy
+Ah, yes. I have predicted your arrival a long time ago, young <PLAYER>. However, your fate is ultimately in your own hands.;Wizard
+Ah, what do we have here? A rare and potent elixir, it seems. Yes, this will make a fine addition to my reagent chest.;Wizard
+Ah... an arctic shard. A most powerful substance, indeed, and potently infused with the element of ice. I'll take it, if you don't mind. ... ...Erm...You're still here. Oh. You want something? ...Take this.;Wizard
+Hmm... I suppose I could accompany you... yes. Once you've arrived, I shall appear. Farewell.;Wizard
+Though the unusual rains may be alarming, there is nothing to fear on this day. In fact, it is a day of great joy for many living things.;Wizard
+Nature is a powerful force, not to be trifled with. Excuse me, but there are special matters I can only attend to on this day.;Wizard
+Marriage... hmm? I pray that you have better luck than I did...;Wizard
+It will be a challenge, to live with one such as Krobus. But richly rewarding to both parties, I'm sure. I admit, I am most interested to see the outcome of such an unusual friendship.;Wizard
+Ah, yes. I have seen many things about your future, young <PLAYER>. If I were to tell you, there could be grave consequences.;Wizard
+Outside the domain of human civilization is where nature truly reveals its secrets.;Wizard
+There are areas of the world, yet unsullied by the constraints of modern thinking, where magic is still very much alive...;Wizard
+While the scientific practice may accurately describe the material of our world, it will never explain the deepest mysteries...;Wizard
+If you have nothing important to tell me, leave me be. I have much work to do.;Wizard
+I trust you enough to allow you in my basement. Treat what you find there with respect.;Wizard
+Greetings, <PLAYER>. I trust you haven't been dabbling in the dark arts?;Wizard
+If reality is constructed by the mind, then if one were to master the power of mind, they could control reality...;Wizard
+It takes years of study to understand the language of the elementals. To actually speak their language requires a lifetime of devoted effort. Now, if you'll excuse me...;Wizard
+There are many mysteries around us. You must be patient if you wish to discover them.;Wizard
+You musn't tell anyone... but I have reason to believe that one of the locals is actually my daughter... I could be mistaken... It's rare, but it does happen.;Wizard
+I believe the townsfolk are afraid of me. It is unfortunate, but I suppose it is human to be afraid of the unknown.;Wizard
+I was married, long ago. After my wife and I split, she began flying around the countryside, cursing everything she saw. It's not a very mature way to express one's anger.;Wizard
+Beware, you are standing above a potent magical field. I built my hut right here on purpose, you know.;Wizard
+Have you made any headway with the forest spirits?;Wizard
+I sometimes observe the local villagers in secret. I am hoping to find an apprentice. Some day I will leave this mortal plane, but my arcane pursuits must continue.;Wizard
diff --git a/lab1/data/dialogues.xlsx b/lab1/data/dialogues.xlsx
new file mode 100644
index 0000000..0d4ae7e
Binary files /dev/null and b/lab1/data/dialogues.xlsx differ
diff --git a/lab1/data/metadata/characters.csv b/lab1/data/metadata/characters.csv
new file mode 100644
index 0000000..997d37d
--- /dev/null
+++ b/lab1/data/metadata/characters.csv
@@ -0,0 +1,35 @@
+character,gender,age_group,species,role
+Abigail,female,teen,human,дочь Пьера; искательница приключений
+Alex,male,teen,human,спортсмен; сын Джорджа и Эвелины
+Clint,male,adult,human,кузнец
+Demetrius,male,adult,human,учёный; муж Робин
+Dwarf,unknown,adult,monster,подземный торговец
+Elliott,male,adult,human,писатель
+Emily,female,adult,human,швея; работает в салуне
+Evelyn,female,elder,human,бабушка Алекса
+George,male,elder,human,дедушка Алекса
+Gus,male,adult,human,владелец салуна
+Haley,female,teen,human,фотограф; сестра Эмили
+Harvey,male,adult,human,врач
+Jas,female,child,human,племянница Мэрни
+Jodi,female,adult,human,мать Сэма и Винсента
+Kent,male,adult,human,военный; муж Джоди
+Krobus,unknown,adult,monster,теневой житель канализации
+Leah,female,adult,human,художница
+Leo,male,child,human,ребёнок с острова
+Lewis,male,elder,human,мэр
+Linus,male,adult,human,отшельник
+Marnie,female,adult,human,владелица ранчо
+Maru,female,adult,human,изобретательница; дочь Робин и Деметриуса
+Mister Qi,unknown,unknown,unknown,таинственный незнакомец
+Pam,female,adult,human,мать Пенни; водитель автобуса
+Penny,female,adult,human,учительница; дочь Пэм
+Pierre,male,adult,human,владелец магазина
+Robin,female,adult,human,плотник; жена Деметриуса
+Sam,male,teen,human,музыкант; сын Джоди и Кента
+Sandy,female,adult,human,продавщица в пустыне
+Sebastian,male,teen,human,программист; сын Робин
+Shane,male,adult,human,работник ранчо
+Vincent,male,child,human,сын Джоди и Кента
+Willy,male,elder,human,рыбак
+Wizard,male,adult,unknown,волшебник
diff --git a/lab1/data/stats_per_character.xlsx b/lab1/data/stats_per_character.xlsx
new file mode 100644
index 0000000..d386623
Binary files /dev/null and b/lab1/data/stats_per_character.xlsx differ
diff --git a/lab1/experiments/README.md b/lab1/experiments/README.md
new file mode 100644
index 0000000..2ed41d0
--- /dev/null
+++ b/lab1/experiments/README.md
@@ -0,0 +1,18 @@
+# Сценарии экспериментов
+
+Основные сценарии запускаются через командный интерфейс пакета `stylometry`:
+
+```bash
+cd lab1
+python -m stylometry.cli --help
+```
+
+Полный протокол и список экспериментов E0–E17 описаны в `../EXPERIMENTS.md`.
+Готовая последовательность базовых запусков — в
+`../scripts/run_core_experiments.sh`.
+
+Этот каталог оставлен для отдельных сценариев, которые могут понадобиться
+позже: например, агрегация реплик в псевдодокументы по 30 предложений,
+тематический контроль или дополнительные признаки. Такие сценарии должны
+использовать те же модули `stylometry` и сохранять результаты в
+`../results/` по той же схеме.
diff --git a/lab1/main.py b/lab1/main.py
new file mode 100644
index 0000000..3557890
--- /dev/null
+++ b/lab1/main.py
@@ -0,0 +1,16 @@
+"""Точка входа для запуска команд стилометрии.
+
+Пример:
+    python main.py audit
+    python main.py run --preset stylometry_pos_lsa --model logreg --repeats 5
+"""
+
+from __future__ import annotations
+
+import sys
+
+from stylometry.cli import main
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/lab1/requirements.txt b/lab1/requirements.txt
new file mode 100644
index 0000000..8919260
--- /dev/null
+++ b/lab1/requirements.txt
@@ -0,0 +1,10 @@
+numpy>=2.0,<3
+pandas>=2.2,<4
+scipy>=1.12,<2
+scikit-learn>=1.5,<2
+matplotlib>=3.8,<4
+openpyxl>=3.1,<4
+nltk>=3.9,<4
+joblib>=1.3,<2
+tabulate>=0.9,<1
+pytest>=8.0,<10
diff --git a/lab1/results/README.md b/lab1/results/README.md
new file mode 100644
index 0000000..d9eded6
--- /dev/null
+++ b/lab1/results/README.md
@@ -0,0 +1,30 @@
+# Результаты запусков
+
+Каталоги `*_<timestamp>` создаются командой `python -m stylometry.cli run`.
+В каждом запуске сохраняются:
+
+- `run_metadata.json` — конфигурация, версии библиотек, seed;
+- `fold_metrics.csv` — метрики по блокам перекрёстной проверки;
+- `pooled_metrics.csv` — метрики по объединённым out-of-fold предсказаниям;
+- `bootstrap_ci.json` — доверительные интервалы;
+- `oof_predictions.csv` — предсказания для каждой реплики;
+- `per_class_report.csv` — точность, полнота и F-мера по персонажам;
+- `confusion_matrix.csv` и `confusion_matrix_normalized.csv` — матрицы ошибок;
+- `top_confusions.csv` — самые частые пары ошибок;
+- `summary.md` — краткий отчёт;
+- `figures/` — графики.
+
+Каталоги, которые уже лежат здесь, — **предварительные технические прогоны**,
+сделанные при разработке конвейера. Они не являются финальными результатами:
+часть запусков выполнена с одним повтором и без вложенной настройки
+гиперпараметров. Для отчёта нужно перезапустить эксперименты по протоколу
+`../EXPERIMENTS.md` и использовать свежие каталоги.
+
+Команды `length-report`, `compare` и `subgroup-report` читают сохранённый
+каталог и добавляют в него новые таблицы:
+
+```bash
+python -m stylometry.cli length-report --run-dir results/<каталог>
+python -m stylometry.cli compare --run-a results/<A> --run-b results/<B>
+python -m stylometry.cli subgroup-report --run-dir results/<каталог> --group gender
+```
diff --git a/lab1/scripts/check_environment.py b/lab1/scripts/check_environment.py
new file mode 100644
index 0000000..2e6fa31
--- /dev/null
+++ b/lab1/scripts/check_environment.py
@@ -0,0 +1,77 @@
+"""Проверка окружения перед серией экспериментов."""
+
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from stylometry.io import DEFAULT_DATA_PATH, corpus_fingerprint, read_corpus, validate_corpus  # noqa: E402
+from stylometry.tokenization import Tokenizer  # noqa: E402
+
+
+def package_versions() -> dict[str, str]:
+    versions: dict[str, str] = {}
+    for name in ("numpy", "pandas", "scipy", "sklearn", "nltk", "matplotlib"):
+        try:
+            module = __import__(name)
+            versions[name] = getattr(module, "__version__", "unknown")
+        except Exception:
+            versions[name] = "missing"
+    return versions
+
+
+def main() -> int:
+    print("=== Проверка окружения ===")
+    print(f"Python: {sys.version.split()[0]} ({sys.executable})")
+    print(f"Каталог проекта: {PROJECT_ROOT}")
+
+    versions = package_versions()
+    print("Версии библиотек:")
+    for name, version in versions.items():
+        print(f"  {name}: {version}")
+    missing = [name for name, version in versions.items() if version == "missing"]
+    if missing:
+        print(f"ОШИБКА: не установлены библиотеки: {missing}")
+        print("Выполните: pip install -r requirements.txt")
+        return 1
+
+    if not DEFAULT_DATA_PATH.exists():
+        print(f"ОШИБКА: нет файла корпуса {DEFAULT_DATA_PATH}")
+        return 1
+
+    frame = read_corpus(DEFAULT_DATA_PATH)
+    stats = validate_corpus(frame)
+    print("Корпус:")
+    print(f"  файл: {DEFAULT_DATA_PATH}")
+    print(f"  SHA-256: {corpus_fingerprint(DEFAULT_DATA_PATH)}")
+    print(f"  реплик: {stats['n_lines']}")
+    print(f"  персонажей: {stats['n_characters']}")
+    print(f"  словоупотреблений: {stats['n_words']}")
+
+    try:
+        tokenizer = Tokenizer(use_nltk=True, strict=True)
+    except Exception as error:
+        print(f"ОШИБКА NLTK: {error}")
+        print("Выполните: python scripts/download_nltk_data.py")
+        return 1
+    analyzed = tokenizer.analyze(frame.iloc[0]["text"])
+    print("NLTK:")
+    print(f"  токенизатор: {'NLTK' if tokenizer.use_nltk else 'резервный'}")
+    print(f"  POS-теггер: {'доступен' if tokenizer.pos_available else 'недоступен'}")
+    print(f"  пример токенов: {analyzed.tokens[:8]}")
+    print(f"  пример POS: {analyzed.upos[:8]}")
+    if not tokenizer.pos_available:
+        print("ОШИБКА: POS-теггер недоступен; эксперименты с POS будут неполными.")
+        return 1
+
+    print("\nОкружение готово.")
+    print(json.dumps({"versions": versions, "n_lines": stats["n_lines"]}, ensure_ascii=False))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/lab1/scripts/collect_results.sh b/lab1/scripts/collect_results.sh
new file mode 100755
index 0000000..0604d66
--- /dev/null
+++ b/lab1/scripts/collect_results.sh
@@ -0,0 +1,56 @@
+#!/usr/bin/env bash
+# Упаковка результатов для передачи коллеге.
+#
+# Запуск из каталога lab1:
+#   bash scripts/collect_results.sh
+#
+# Создаёт exports/nlp_lab1_results_<timestamp>.tar.gz и файл SHA-256.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+cd "${PROJECT_ROOT}"
+
+PYTHON="${PYTHON:-python}"
+RUN_ID="${RUN_ID:-$(date +%Y%m%d_%H%M%S)}"
+OUT_DIR="${OUT_DIR:-${PROJECT_ROOT}/exports}"
+ARCHIVE="${OUT_DIR}/nlp_lab1_results_${RUN_ID}.tar.gz"
+MANIFEST="${OUT_DIR}/nlp_lab1_results_${RUN_ID}_manifest.txt"
+
+mkdir -p "${OUT_DIR}"
+
+{
+    echo "Дата: $(date)"
+    echo "Каталог: ${PROJECT_ROOT}"
+    echo "Интерпретатор: ${PYTHON}"
+    echo
+    "${PYTHON}" scripts/check_environment.py || true
+    echo
+    if git rev-parse --is-inside-work-tree >/dev/null 2>&1; then
+        echo "Git commit: $(git rev-parse HEAD 2>/dev/null || echo 'нет коммита')"
+    else
+        echo "Git: репозиторий не инициализирован"
+    fi
+    echo
+    echo "Содержимое results:"
+    find results -maxdepth 2 -type f | sort | head -n 500
+} > "${MANIFEST}"
+
+INCLUDE=()
+for path in results features data/processed data/metadata; do
+    if [[ -e "${path}" ]]; then
+        INCLUDE+=("${path}")
+    fi
+done
+INCLUDE+=("${MANIFEST#"${PROJECT_ROOT}/"}")
+
+tar -czf "${ARCHIVE}" "${INCLUDE[@]}"
+
+sha256sum "${ARCHIVE}" > "${ARCHIVE}.sha256"
+
+echo "Архив: ${ARCHIVE}"
+echo "Контрольная сумма: ${ARCHIVE}.sha256"
+echo "Манифест: ${MANIFEST}"
+echo
+echo "Передайте коллеге архив и файл .sha256."
diff --git a/lab1/scripts/download_nltk_data.py b/lab1/scripts/download_nltk_data.py
new file mode 100644
index 0000000..badcc8d
--- /dev/null
+++ b/lab1/scripts/download_nltk_data.py
@@ -0,0 +1,29 @@
+"""Скачивает ресурсы NLTK в локальный каталог lab1/nltk_data."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import nltk
+
+PACKAGES = [
+    "punkt",
+    "punkt_tab",
+    "averaged_perceptron_tagger",
+    "averaged_perceptron_tagger_eng",
+    "stopwords",
+]
+
+
+def main() -> int:
+    target = Path(__file__).resolve().parents[1] / "nltk_data"
+    target.mkdir(parents=True, exist_ok=True)
+    for package in PACKAGES:
+        print(f"Скачивание {package} ...")
+        nltk.download(package, download_dir=str(target), quiet=False)
+    print(f"Готово: {target}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/lab1/scripts/run_all_experiments.sh b/lab1/scripts/run_all_experiments.sh
new file mode 100755
index 0000000..f3a52d6
--- /dev/null
+++ b/lab1/scripts/run_all_experiments.sh
@@ -0,0 +1,169 @@
+#!/usr/bin/env bash
+# Полная серия экспериментов для лабораторной работы.
+#
+# Запуск из каталога lab1:
+#   bash scripts/run_all_experiments.sh
+#
+# С другим интерпретатором:
+#   PYTHON=../.venv/bin/python bash scripts/run_all_experiments.sh
+#
+# Быстрая проверка (1 повтор, без вложенной кросс-валидации):
+#   QUICK=1 bash scripts/run_all_experiments.sh
+#
+# Все команды пишутся в results/logs/run_all_<timestamp>.log.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+cd "${PROJECT_ROOT}"
+
+PYTHON="${PYTHON:-python}"
+if [[ "${PYTHON}" == */* ]]; then
+    PYTHON_DIR="$(cd "$(dirname "${PYTHON}")" && pwd)"
+    PYTHON="${PYTHON_DIR}/$(basename "${PYTHON}")"
+fi
+
+if ! "${PYTHON}" -c "import sklearn, pandas, nltk" >/dev/null 2>&1; then
+    echo "ОШИБКА: в интерпретаторе ${PYTHON} не установлены зависимости."
+    echo "Выполните: ${PYTHON} -m pip install -r requirements.txt"
+    exit 1
+fi
+
+REPEATS_MAIN="${REPEATS_MAIN:-5}"
+REPEATS_FAST="${REPEATS_FAST:-3}"
+REPEATS_LSA="${REPEATS_LSA:-3}"
+REPEATS_ABLATE="${REPEATS_ABLATE:-3}"
+REPEATS_DELTA="${REPEATS_DELTA:-2}"
+SKIP_NESTED="${SKIP_NESTED:-0}"
+
+if [[ "${QUICK:-0}" == "1" ]]; then
+    REPEATS_MAIN=1
+    REPEATS_FAST=1
+    REPEATS_LSA=1
+    REPEATS_ABLATE=1
+    REPEATS_DELTA=1
+    SKIP_NESTED=1
+fi
+
+LOG_DIR="${PROJECT_ROOT}/results/logs"
+mkdir -p "${LOG_DIR}"
+RUN_ID="${RUN_ID:-$(date +%Y%m%d_%H%M%S)}"
+LOG_FILE="${LOG_DIR}/run_all_${RUN_ID}.log"
+exec > >(tee -a "${LOG_FILE}") 2>&1
+
+step() {
+    echo
+    echo "============================================================"
+    echo "== $*"
+    echo "============================================================"
+}
+
+run() {
+    echo
+    echo ">>> $*"
+    "$@"
+}
+
+latest_dir() {
+    local prefix="$1"
+    ls -1dt results/"${prefix}"_* 2>/dev/null | head -n 1 || true
+}
+
+echo "Начало серии: $(date)"
+echo "Интерпретатор: ${PYTHON}"
+echo "Повторы: main=${REPEATS_MAIN}, fast=${REPEATS_FAST}, lsa=${REPEATS_LSA}, ablate=${REPEATS_ABLATE}, delta=${REPEATS_DELTA}"
+echo "Вложенная кросс-валидация: ${SKIP_NESTED} (1 — пропустить)"
+echo "Лог: ${LOG_FILE}"
+
+step "0. Проверка окружения и тесты"
+run "${PYTHON}" scripts/check_environment.py
+run "${PYTHON}" -m pytest -q
+
+step "1. Аудит корпуса и ручные признаки (E0)"
+run "${PYTHON}" -m stylometry.cli audit
+run "${PYTHON}" -m stylometry.cli features
+
+step "2. Нижние границы (E1)"
+run "${PYTHON}" -m stylometry.cli run --preset length --model dummy --repeats 1
+run "${PYTHON}" -m stylometry.cli run --preset length --model logreg --repeats "${REPEATS_MAIN}"
+run "${PYTHON}" -m stylometry.cli run --preset char --model multinomial_nb --repeats "${REPEATS_FAST}"
+run "${PYTHON}" -m stylometry.cli run --preset pos --model multinomial_nb --repeats "${REPEATS_FAST}"
+run "${PYTHON}" -m stylometry.cli run --preset fw --model logreg --repeats "${REPEATS_FAST}"
+
+step "3. Классическая стилометрия (E2)"
+run "${PYTHON}" -m stylometry.cli run --preset stylometry --model logreg --repeats "${REPEATS_MAIN}"
+run "${PYTHON}" -m stylometry.cli run --preset stylometry --model linear_svc --repeats "${REPEATS_MAIN}"
+
+step "4. Ручные признаки + n-граммы (E4)"
+run "${PYTHON}" -m stylometry.cli run --preset stylometry_pos --model logreg --repeats "${REPEATS_MAIN}"
+
+step "5. Основной объединённый набор (E8)"
+run "${PYTHON}" -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --repeats "${REPEATS_MAIN}"
+
+step "6. Варианты LSA по отдельности (E5, E7)"
+for variant in word word2 content fw char pos; do
+    run "${PYTHON}" -m stylometry.cli run --preset "lsa_${variant}" --model logreg --repeats "${REPEATS_LSA}"
+done
+
+step "7. Перебор числа компонент LSA (E5)"
+run "${PYTHON}" -m stylometry.cli lsa-sweep \
+    --variants word,content,fw,char,pos \
+    --components 50,100,200 \
+    --model logreg --repeats 1
+
+step "8. Сравнение классификаторов (E6)"
+run "${PYTHON}" -m stylometry.cli benchmark \
+    --preset stylometry_pos_lsa \
+    --models logreg,linear_svc,random_forest,multinomial_nb,knn \
+    --repeats "${REPEATS_FAST}"
+
+step "9. Поочерёдное исключение групп признаков (E3)"
+run "${PYTHON}" -m stylometry.cli ablate \
+    --preset stylometry_pos_lsa --model logreg --repeats "${REPEATS_ABLATE}"
+
+step "10. Дельта Барроуза (E11)"
+run "${PYTHON}" -m stylometry.cli delta \
+    --features 100,200,500 --metrics manhattan,cosine --repeats "${REPEATS_DELTA}"
+
+step "11. Верификация «свой/чужой» (E12)"
+run "${PYTHON}" -m stylometry.cli verify --preset stylometry_pos_lsa
+
+step "12. Кластеризация персонажей (E13)"
+run "${PYTHON}" -m stylometry.cli cluster --preset stylometry_pos_lsa
+
+if [[ "${SKIP_NESTED}" != "1" ]]; then
+    step "13. Вложенная кросс-валидация (E16)"
+    run "${PYTHON}" -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --nested
+    run "${PYTHON}" -m stylometry.cli run --preset stylometry_pos_lsa --model linear_svc --nested
+else
+    echo
+    echo "Вложенная кросс-валидация пропущена (SKIP_NESTED=1 или QUICK=1)."
+fi
+
+step "14. Анализ длины, подгрупп и сравнение запусков (E9, E14, E15)"
+MAIN_DIR="$(latest_dir stylometry_pos_lsa_logreg)"
+POS_DIR="$(latest_dir stylometry_pos_logreg)"
+if [[ -n "${MAIN_DIR}" ]]; then
+    run "${PYTHON}" -m stylometry.cli length-report --run-dir "${MAIN_DIR}"
+    for group in gender age_group species; do
+        run "${PYTHON}" -m stylometry.cli subgroup-report \
+            --run-dir "${MAIN_DIR}" --group "${group}"
+    done
+    if [[ -n "${POS_DIR}" ]]; then
+        run "${PYTHON}" -m stylometry.cli compare \
+            --run-a "${MAIN_DIR}" --run-b "${POS_DIR}" \
+            --out "${MAIN_DIR}/comparison_main_vs_pos.json"
+    fi
+else
+    echo "Не найден каталог основного запуска; анализ пропущен."
+fi
+
+step "Готово"
+echo "Конец серии: $(date)"
+echo
+echo "Основные каталоги результатов:"
+ls -1dt results/*/ 2>/dev/null | head -n 30 || true
+echo
+echo "Полный лог: ${LOG_FILE}"
+echo "Для упаковки результатов: bash scripts/collect_results.sh"
diff --git a/lab1/scripts/run_core_experiments.sh b/lab1/scripts/run_core_experiments.sh
new file mode 100755
index 0000000..7121b13
--- /dev/null
+++ b/lab1/scripts/run_core_experiments.sh
@@ -0,0 +1,31 @@
+#!/usr/bin/env bash
+# Базовая серия экспериментов. Запускать из каталога lab1.
+# Для финального отчёта увеличьте --repeats до 5.
+
+set -euo pipefail
+
+python -m stylometry.cli audit
+python -m stylometry.cli features
+
+# E1: нижние границы
+python -m stylometry.cli run --preset length --model dummy --repeats 1
+python -m stylometry.cli run --preset length --model logreg --repeats 3
+python -m stylometry.cli run --preset char --model multinomial_nb --repeats 3
+python -m stylometry.cli run --preset pos --model multinomial_nb --repeats 3
+python -m stylometry.cli run --preset fw --model logreg --repeats 3
+
+# E2-E4: ручные признаки, POS и объединение
+python -m stylometry.cli run --preset stylometry --model logreg --repeats 3
+python -m stylometry.cli run --preset stylometry_pos --model logreg --repeats 3
+python -m stylometry.cli run --preset stylometry_pos_lsa --model logreg --repeats 3
+python -m stylometry.cli ablate --preset stylometry_pos_lsa --model logreg --repeats 2
+
+# E5-E6: LSA и классификаторы
+python -m stylometry.cli lsa-sweep --variants word,pos,char --components 50,100,200 --model logreg --repeats 1
+python -m stylometry.cli benchmark --preset stylometry_pos_lsa \
+    --models logreg,linear_svc,random_forest,multinomial_nb,knn --repeats 2
+
+# E11-E13: дельта, верификация, кластеризация
+python -m stylometry.cli delta --features 100,200,500 --metrics manhattan,cosine --repeats 2
+python -m stylometry.cli verify --preset stylometry_pos_lsa
+python -m stylometry.cli cluster --preset stylometry_pos_lsa
diff --git a/lab1/stylometry/__init__.py b/lab1/stylometry/__init__.py
new file mode 100644
index 0000000..119019b
--- /dev/null
+++ b/lab1/stylometry/__init__.py
@@ -0,0 +1,13 @@
+"""Стилометрия реплик Stardew Valley.
+
+Основные модули:
+- ``io`` — чтение и валидация корпуса;
+- ``cleaning`` — нормализация текста;
+- ``tokenization`` — NLTK-токенизация, Universal POS, резервный режим;
+- ``features`` — ручные стилометрические признаки;
+- ``lsa`` — TF-IDF и латентно-семантический анализ;
+- ``pipelines`` — сборка признаковых пространств и классификаторов;
+- ``cv``, ``evaluation``, ``stats`` — кросс-валидация, метрики и статистика.
+"""
+
+__version__ = "1.0.0"
diff --git a/lab1/stylometry/cleaning.py b/lab1/stylometry/cleaning.py
new file mode 100644
index 0000000..e558dd2
--- /dev/null
+++ b/lab1/stylometry/cleaning.py
@@ -0,0 +1,123 @@
+"""Нормализация текста реплик.
+
+Модуль намеренно не выполняет агрессивную очистку: корпус уже отфильтрован
+вручную. Здесь только приводятся к единому виду Unicode, плейсхолдеры и
+пробельные символы. Пунктуация, регистр и разговорные элементы сохраняются,
+поскольку они являются стилометрическими признаками.
+"""
+
+from __future__ import annotations
+
+import re
+import unicodedata
+from typing import Iterable
+
+# Плейсхолдеры игрового движка. Они не являются авторским текстом, поэтому
+# заменяются на нейтральные слова, а их количество учитывается отдельным
+# признаком.
+PLACEHOLDER_PATTERN = re.compile(r"<([A-Z][A-Z0-9_]*)>")
+
+PLACEHOLDER_MAP: dict[str, str] = {
+    "PLAYER": "PLAYER",
+    "FARM": "FARM",
+    "PET": "PET",
+    "SPOUSE": "SPOUSE",
+    "TIME": "TIME",
+    "NOUN": "NOUN",
+    "FIRST_LETTER": "FIRST_LETTER",
+    "NAME": "NAME",
+}
+
+# Символы, которые в корпусе могут встречаться в разных типографских
+# вариантах. Приводим их к ASCII-эквивалентам, чтобы признаки были
+# сопоставимы между репликами.
+_UNICODE_REPLACEMENTS: dict[str, str] = {
+    "\u2026": "...",  # многоточие
+    "\u2018": "'",
+    "\u2019": "'",
+    "\u201a": "'",
+    "\u201b": "'",
+    "\u201c": '"',
+    "\u201d": '"',
+    "\u201e": '"',
+    "\u2013": "-",  # короткое тире
+    "\u2014": "-",  # длинное тире
+    "\u2212": "-",  # математический минус
+    "\u00a0": " ",  # неразрывный пробел
+    "\u2009": " ",  # тонкий пробел
+    "\u200a": " ",
+    "\u200b": "",   # нулевой пробел
+    "\u200c": "",
+    "\u200d": "",
+    "\ufeff": "",
+}
+
+
+def normalize_unicode(text: str) -> str:
+    """Приводит Unicode к предсказуемому виду, не меняя смысл текста."""
+    text = unicodedata.normalize("NFKC", str(text))
+    for source, target in _UNICODE_REPLACEMENTS.items():
+        text = text.replace(source, target)
+    return text
+
+
+def normalize_placeholders(text: str) -> str:
+    """Заменяет ``<PLAYER>`` и аналогичные метки на слова без угловых скобок."""
+
+    def _replace(match: re.Match[str]) -> str:
+        name = match.group(1)
+        return PLACEHOLDER_MAP.get(name, "ENTITY")
+
+    return PLACEHOLDER_PATTERN.sub(_replace, text)
+
+
+def normalize_whitespace(text: str) -> str:
+    """Схлопывает повторные пробелы и переводы строк."""
+    text = re.sub(r"[ \t\f\v]+", " ", text)
+    text = re.sub(r"\s*\n\s*", " ", text)
+    return text.strip()
+
+
+def clean_text(text: str) -> str:
+    """Основная функция нормализации одной реплики."""
+    text = normalize_unicode(text)
+    text = normalize_placeholders(text)
+    text = normalize_whitespace(text)
+    return text
+
+
+def normalized_key(text: str) -> str:
+    """Ключ для поиска точных дублей.
+
+    Регистр, пунктуация и плейсхолдеры игнорируются. Такой ключ использовался
+    при аудите корпуса; для обучения он не применяется.
+    """
+    text = text.lower()
+    text = PLACEHOLDER_PATTERN.sub("", text)
+    text = re.sub(r"[^a-z0-9' ]+", " ", text)
+    return normalize_whitespace(text)
+
+
+def count_placeholders(text: str) -> int:
+    """Возвращает число плейсхолдеров в исходной реплике."""
+    return len(PLACEHOLDER_PATTERN.findall(text))
+
+
+def placeholder_types(text: str) -> list[str]:
+    """Возвращает типы плейсхолдеров в порядке появления."""
+    return [match.group(1) for match in PLACEHOLDER_PATTERN.finditer(text)]
+
+
+def is_valid_line(text: str) -> bool:
+    """Отбрасывает пустые строки и строки без латинских букв."""
+    return bool(re.search(r"[A-Za-z]", text))
+
+
+def tokenize_for_audit(text: str) -> list[str]:
+    """Простая токенизация для аудита; не заменяет NLTK-токенизатор."""
+    return re.findall(r"[A-Za-z]+(?:'[A-Za-z]+)?|\d+|[^\w\s]", text)
+
+
+def join_tokens(tokens: Iterable[str]) -> str:
+    """Склеивает токены через пробел, сохраняя их как отдельные единицы."""
+    return " ".join(str(token) for token in tokens)
diff --git a/lab1/stylometry/cli.py b/lab1/stylometry/cli.py
new file mode 100644
index 0000000..5f12e75
--- /dev/null
+++ b/lab1/stylometry/cli.py
@@ -0,0 +1,866 @@
+"""Командный интерфейс для аудита, признаков, LSA и экспериментов."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from datetime import datetime
+from pathlib import Path
+from typing import Any, Sequence
+
+import numpy as np
+import pandas as pd
+
+from .clustering import run_clustering
+from .cv import CVResult, run_nested_cv, run_repeated_cv
+from .datasets import DEFAULT_CACHE_PATH, load_feature_frame
+from .delta import DeltaClassifier
+from .evaluation import (
+    bootstrap_ci,
+    classification_metrics,
+    confusion_matrix_frame,
+    per_class_report,
+    top_confusions,
+)
+from .features import StylometricVectorizer, feature_dictionary
+from .io import (
+    DEFAULT_DATA_PATH,
+    compare_with_stats,
+    corpus_fingerprint,
+    read_corpus,
+    read_stats,
+    validate_corpus,
+    write_csv,
+)
+from .lsa import LSA_VARIANTS, make_lsa_vectorizer
+from .metadata import group_confusions, load_character_metadata, subgroup_report
+from .pipelines import (
+    MODEL_NAMES,
+    FeatureConfig,
+    build_pipeline,
+    preset_config,
+)
+from .plotting import (
+    plot_class_distribution,
+    plot_confusion_matrix,
+    plot_pca,
+    plot_scree,
+)
+from .stats import bootstrap_difference, mcnemar_exact
+from .verification import run_verification
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+DEFAULT_RESULTS_DIR = PROJECT_ROOT / "results"
+DEFAULT_CONFIG_PATH = PROJECT_ROOT / "configs" / "default.toml"
+
+
+def _package_versions() -> dict[str, str]:
+    versions = {}
+    for name in ("numpy", "pandas", "scipy", "sklearn", "nltk", "matplotlib"):
+        try:
+            module = __import__(name)
+            versions[name] = getattr(module, "__version__", "unknown")
+        except Exception:
+            versions[name] = "missing"
+    return versions
+
+
+def _json_default(value: Any) -> Any:
+    if isinstance(value, (np.integer,)):
+        return int(value)
+    if isinstance(value, (np.floating,)):
+        return float(value)
+    if isinstance(value, (np.ndarray,)):
+        return value.tolist()
+    if isinstance(value, Path):
+        return str(value)
+    return str(value)
+
+
+def save_json(data: Any, path: str | Path) -> Path:
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with path.open("w", encoding="utf-8") as stream:
+        json.dump(data, stream, ensure_ascii=False, indent=2, default=_json_default)
+    return path
+
+
+def _run_id(prefix: str) -> str:
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    return f"{prefix}_{timestamp}"
+
+
+def _safe_name(value: str) -> str:
+    return "".join(char if char.isalnum() or char in "-_." else "_" for char in value)
+
+
+def _markdown_table(frame: pd.DataFrame) -> str:
+    try:
+        return frame.to_markdown(index=False)
+    except Exception:
+        return "```\n" + frame.to_string(index=False) + "\n```"
+
+
+def _load_config(path: str | Path | None) -> dict[str, Any]:
+    if path is None:
+        path = DEFAULT_CONFIG_PATH
+    path = Path(path)
+    if not path.exists():
+        return {}
+    try:
+        import tomllib
+
+        with path.open("rb") as stream:
+            return tomllib.load(stream)
+    except Exception as error:
+        print(f"[config] не удалось прочитать {path}: {error}")
+        return {}
+
+
+def _feature_config(preset: str, overrides: dict[str, Any] | None = None) -> FeatureConfig:
+    config = preset_config(preset)
+    if overrides:
+        for key, value in overrides.items():
+            if hasattr(config, key):
+                setattr(config, key, value)
+    return config
+
+
+def _frame(args: argparse.Namespace) -> pd.DataFrame:
+    data_path = Path(args.data) if getattr(args, "data", None) else DEFAULT_DATA_PATH
+    cache_path = Path(args.cache) if getattr(args, "cache", None) else DEFAULT_CACHE_PATH
+    return load_feature_frame(
+        data_path=data_path,
+        cache_path=cache_path,
+        force=getattr(args, "no_cache", False),
+        verbose=True,
+    )
+
+
+def _first_repeat(oof: pd.DataFrame) -> pd.DataFrame:
+    if oof.empty:
+        return oof
+    repeat = oof["repeat"].min()
+    return oof[oof["repeat"] == repeat].copy()
+
+
+def _save_cv_result(
+    result: CVResult,
+    output_dir: Path,
+    config: dict[str, Any],
+    title: str,
+) -> pd.DataFrame:
+    output_dir.mkdir(parents=True, exist_ok=True)
+    config.setdefault("package_versions", _package_versions())
+    result.fold_metrics.to_csv(output_dir / "fold_metrics.csv", index=False, encoding="utf-8-sig")
+    result.oof.to_csv(output_dir / "oof_predictions.csv", index=False, encoding="utf-8-sig")
+
+    fold_summary = result.summary()
+    fold_summary.to_csv(output_dir / "fold_summary.csv", index=False, encoding="utf-8-sig")
+
+    pooled, bootstrap = result.pooled_summary()
+    pooled.to_csv(output_dir / "pooled_metrics.csv", index=False, encoding="utf-8-sig")
+    save_json(bootstrap, output_dir / "bootstrap_ci.json")
+
+    first = _first_repeat(result.oof)
+    per_class = per_class_report(first["y_true"], first["y_pred"], result.labels)
+    per_class.to_csv(output_dir / "per_class_report.csv", index=False, encoding="utf-8-sig")
+    matrix = confusion_matrix_frame(first["y_true"], first["y_pred"], result.labels)
+    matrix.to_csv(output_dir / "confusion_matrix.csv", encoding="utf-8-sig")
+    normalized = confusion_matrix_frame(
+        first["y_true"], first["y_pred"], result.labels, normalize="true"
+    )
+    normalized.to_csv(output_dir / "confusion_matrix_normalized.csv", encoding="utf-8-sig")
+    top_confusions(first["y_true"], first["y_pred"], result.labels).to_csv(
+        output_dir / "top_confusions.csv", index=False, encoding="utf-8-sig"
+    )
+    plot_confusion_matrix(
+        matrix, output_dir / "figures" / "confusion_matrix.png", title="Матрица ошибок"
+    )
+    test_distribution = pd.DataFrame({"character": first["y_true"].values})
+    plot_class_distribution(
+        test_distribution,
+        character_column="character",
+        path=output_dir / "figures" / "test_class_distribution.png",
+    )
+
+    summary_row = pooled.iloc[0].to_dict() if not pooled.empty else {}
+    summary_row.update(
+        {
+            "model": config.get("model", ""),
+            "preset": config.get("preset", ""),
+            "folds": result.n_splits,
+            "repeats": result.n_repeats,
+            "n_samples": int(len(result.oof) // max(result.n_repeats, 1)),
+            "n_classes": int(len(result.labels)),
+        }
+    )
+    save_json({"title": title, "config": config, "metrics": summary_row, "bootstrap": bootstrap}, output_dir / "run_metadata.json")
+
+    lines = [
+        f"# {title}",
+        "",
+        f"- Модель: `{config.get('model', '')}`",
+        f"- Набор признаков: `{config.get('preset', '')}`",
+        f"- Кросс-валидация: {result.n_splits}-fold, повторов: {result.n_repeats}",
+        f"- Объектов: {summary_row.get('n_samples', '')}",
+        f"- Классов: {summary_row.get('n_classes', '')}",
+        "",
+        "## Основные метрики",
+        "",
+        _markdown_table(fold_summary),
+        "",
+        "## Pooled out-of-fold",
+        "",
+        _markdown_table(pooled),
+        "",
+        "## Доверительные интервалы (bootstrap)",
+        "",
+        "```json",
+        json.dumps(bootstrap, ensure_ascii=False, indent=2, default=_json_default),
+        "```",
+        "",
+    ]
+    (output_dir / "summary.md").write_text("\n".join(lines), encoding="utf-8")
+    return fold_summary
+
+
+def cmd_audit(args: argparse.Namespace) -> int:
+    data_path = Path(args.data) if args.data else DEFAULT_DATA_PATH
+    frame = read_corpus(data_path)
+    stats = validate_corpus(frame)
+    counts = pd.DataFrame(
+        sorted(stats["class_counts"].items(), key=lambda item: item[1], reverse=True),
+        columns=["Персонаж", "Реплик"],
+    )
+    word_lengths = frame["text"].str.findall(r"[A-Za-z0-9']+").map(len)
+    bins = pd.cut(
+        word_lengths,
+        bins=[0, 5, 10, 15, 20, 30, 1000],
+        labels=["0–5", "6–10", "11–15", "16–20", "21–30", "31+"],
+        include_lowest=True,
+    ).value_counts().sort_index()
+    bin_frame = pd.DataFrame({"Длина реплики (слов)": bins.index.astype(str), "Реплик": bins.values})
+
+    comparison = compare_with_stats(frame, read_stats())
+    output_dir = Path(args.out) if args.out else PROJECT_ROOT / "data" / "processed"
+    output_dir.mkdir(parents=True, exist_ok=True)
+    write_csv(frame, output_dir / "lines.csv")
+
+    lines = [
+        "# Аудит корпуса",
+        "",
+        f"- Файл: `{data_path}`",
+        f"- SHA-256: `{corpus_fingerprint(data_path)}`",
+        f"- Реплик: {stats['n_lines']}",
+        f"- Персонажей: {stats['n_characters']}",
+        f"- Словоупотреблений: {stats['n_words']}",
+        f"- Средняя длина: {stats['mean_words_per_line']:.2f} слова",
+        f"- Медиана: {stats['median_words_per_line']:.1f} слова",
+        f"- Минимум/максимум: {stats['min_words_per_line']} / {stats['max_words_per_line']} слов",
+        f"- Точных дублей текста: {stats['n_duplicate_texts']}",
+        f"- Реплик с плейсхолдерами: {stats['n_placeholder_lines']}",
+        f"- Всего плейсхолдеров: {stats['n_placeholders']}",
+        "",
+        "## Распределение по персонажам",
+        "",
+        _markdown_table(counts),
+        "",
+        "## Распределение длин реплик",
+        "",
+        _markdown_table(bin_frame),
+        "",
+    ]
+    if not comparison.empty:
+        lines.extend(
+            [
+                "## Сверка с stats_per_character.xlsx",
+                "",
+                _markdown_table(comparison.reset_index()),
+                "",
+            ]
+        )
+    (output_dir / "audit.md").write_text("\n".join(lines), encoding="utf-8")
+    plot_class_distribution(frame, path=output_dir / "figures" / "class_distribution.png")
+
+    print(f"[audit] реплик: {stats['n_lines']}, персонажей: {stats['n_characters']}")
+    print(f"[audit] отчёт: {output_dir / 'audit.md'}")
+    return 0
+
+
+def cmd_features(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    output_dir = Path(args.out) if args.out else PROJECT_ROOT / "features"
+    output_dir.mkdir(parents=True, exist_ok=True)
+    vectorizer = StylometricVectorizer()
+    matrix = vectorizer.fit_transform(frame)
+    names = vectorizer.get_feature_names_out()
+    result = pd.DataFrame(matrix, columns=names)
+    result.insert(0, "character", frame["character"].values)
+    result.insert(0, "line_id", frame["line_id"].values)
+    write_csv(result, output_dir / "stylometric.csv")
+    dictionary = feature_dictionary(names)
+    write_csv(dictionary, output_dir / "feature_dictionary.csv")
+    print(f"[features] матрица: {matrix.shape}; словарь: {output_dir / 'feature_dictionary.csv'}")
+    return 0
+
+
+def cmd_lsa(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    output_dir = Path(args.out) if args.out else PROJECT_ROOT / "features"
+    output_dir.mkdir(parents=True, exist_ok=True)
+    variant = args.variant
+    components = args.components
+    vectorizer = make_lsa_vectorizer(variant, n_components=components)
+    matrix = vectorizer.fit_transform(frame)
+    result = pd.DataFrame(matrix, columns=vectorizer.get_feature_names_out())
+    result.insert(0, "character", frame["character"].values)
+    result.insert(0, "line_id", frame["line_id"].values)
+    path = output_dir / f"lsa_{variant}_k{vectorizer.n_components_}.csv"
+    write_csv(result, path)
+
+    top_terms = []
+    for component_index, terms in enumerate(vectorizer.top_terms(15), start=1):
+        for rank, (term, weight) in enumerate(terms, start=1):
+            top_terms.append(
+                {
+                    "component": component_index,
+                    "rank": rank,
+                    "term": term,
+                    "weight": weight,
+                }
+            )
+    write_csv(pd.DataFrame(top_terms), output_dir / f"lsa_{variant}_top_terms.csv")
+    variance = pd.DataFrame(
+        {
+            "component": np.arange(1, vectorizer.n_components_ + 1),
+            "explained_variance_ratio": vectorizer.explained_variance_ratio_,
+            "cumulative_explained_variance": vectorizer.cumulative_explained_variance_,
+        }
+    )
+    write_csv(variance, output_dir / f"lsa_{variant}_variance.csv")
+    plot_scree(
+        vectorizer.explained_variance_ratio_,
+        output_dir / f"lsa_{variant}_scree.png",
+        title=f"LSA ({variant}): объяснённая дисперсия",
+    )
+    print(
+        f"[lsa] вариант={variant}, компонент={vectorizer.n_components_}, "
+        f"накопленная дисперсия={vectorizer.cumulative_explained_variance_[-1]:.3f}"
+    )
+    print(f"[lsa] матрица: {path}")
+    return 0
+
+
+def cmd_run(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    y = frame["character"].values
+    config_data = _load_config(args.config)
+    preset = args.preset or config_data.get("experiment", {}).get("preset", "stylometry_pos_lsa")
+    model = args.model or config_data.get("experiment", {}).get("model", "logreg")
+    folds = (
+        args.folds if args.folds is not None
+        else config_data.get("cv", {}).get("folds", 5)
+    )
+    repeats = (
+        args.repeats if args.repeats is not None
+        else config_data.get("cv", {}).get("repeats", 5)
+    )
+    seed = (
+        args.seed if args.seed is not None
+        else config_data.get("cv", {}).get("seed", 42)
+    )
+
+    overrides = dict(config_data.get("features", {}))
+    for key in ("char_ngram_range", "pos_ngram_range"):
+        if key in overrides and isinstance(overrides[key], list):
+            overrides[key] = tuple(overrides[key])
+    if "lsa_variants" in overrides and isinstance(overrides["lsa_variants"], list):
+        overrides["lsa_variants"] = tuple(overrides["lsa_variants"])
+    feature_config = _feature_config(preset, overrides)
+    pipeline = build_pipeline(feature_config, model)
+    run_config = {
+        "preset": preset,
+        "model": model,
+        "folds": folds,
+        "repeats": repeats,
+        "seed": seed,
+        "n_samples": int(len(frame)),
+        "n_classes": int(len(np.unique(y))),
+        "corpus_sha256": frame.attrs.get("corpus_sha256", ""),
+        "feature_config": dict(feature_config.__dict__),
+        "package_versions": _package_versions(),
+    }
+
+    if args.nested:
+        result = run_nested_cv(
+            pipeline, frame, y, model_name=model,
+            outer_splits=folds, inner_splits=3, seed=seed, verbose=True,
+        )
+        output_dir = DEFAULT_RESULTS_DIR / _run_id(f"nested_{preset}_{model}")
+        _save_cv_result(result, output_dir, run_config, f"Вложенная CV: {preset} / {model}")
+        save_json(result.best_params, output_dir / "best_params.json")
+    else:
+        result = run_repeated_cv(
+            pipeline, frame, y, n_splits=folds, n_repeats=repeats,
+            seed=seed, model_name=model, verbose=True,
+        )
+        output_dir = DEFAULT_RESULTS_DIR / _run_id(f"{preset}_{model}")
+        _save_cv_result(result, output_dir, run_config, f"CV: {preset} / {model}")
+
+    print(f"[run] результаты: {output_dir}")
+    print(result.summary().to_string(index=False))
+    return 0
+
+
+def cmd_benchmark(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    y = frame["character"].values
+    preset = args.preset
+    models = args.models.split(",") if args.models else [
+        "logreg", "linear_svc", "random_forest", "multinomial_nb", "knn"
+    ]
+    output_dir = DEFAULT_RESULTS_DIR / _run_id(f"benchmark_{preset}")
+    output_dir.mkdir(parents=True, exist_ok=True)
+    rows = []
+    for model in models:
+        model = model.strip()
+        if model not in MODEL_NAMES:
+            print(f"[benchmark] пропускаю неизвестную модель {model}")
+            continue
+        print(f"\n[benchmark] модель: {model}")
+        pipeline = build_pipeline(_feature_config(preset), model)
+        result = run_repeated_cv(
+            pipeline, frame, y, n_splits=args.folds, n_repeats=args.repeats,
+            seed=args.seed, model_name=model, verbose=False,
+        )
+        pooled, bootstrap = result.pooled_summary()
+        first = pooled.iloc[0].to_dict() if not pooled.empty else {}
+        rows.append(
+            {
+                "model": model,
+                "macro_f1": first.get("macro_f1"),
+                "balanced_accuracy": first.get("balanced_accuracy"),
+                "accuracy": first.get("accuracy"),
+                "mcc": first.get("mcc"),
+                "cohen_kappa": first.get("cohen_kappa"),
+                "top3_accuracy": first.get("top3_accuracy"),
+                "top5_accuracy": first.get("top5_accuracy"),
+                "macro_f1_ci_lower": bootstrap["macro_f1"]["lower"],
+                "macro_f1_ci_upper": bootstrap["macro_f1"]["upper"],
+            }
+        )
+        model_dir = output_dir / _safe_name(model)
+        _save_cv_result(
+            result,
+            model_dir,
+            {
+                "preset": preset,
+                "model": model,
+                "corpus_sha256": frame.attrs.get("corpus_sha256", ""),
+            },
+            f"Benchmark: {preset} / {model}",
+        )
+    summary = pd.DataFrame(rows).sort_values("macro_f1", ascending=False)
+    write_csv(summary, output_dir / "summary.csv")
+    print("\n" + summary.to_string(index=False))
+    return 0
+
+
+def cmd_ablate(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    y = frame["character"].values
+    base = _feature_config(args.preset)
+    groups = list(base.groups)
+    if not groups:
+        print("[ablate] в выбранном наборе нет ручных групп признаков.")
+        return 1
+    output_dir = DEFAULT_RESULTS_DIR / _run_id(f"ablation_{args.preset}")
+    output_dir.mkdir(parents=True, exist_ok=True)
+    rows = []
+    variants = [("full", groups)] + [
+        (f"without_{group}", [value for value in groups if value != group])
+        for group in groups
+    ]
+    for name, selected in variants:
+        config = FeatureConfig(
+            groups=tuple(selected),
+            use_char_ngrams=base.use_char_ngrams,
+            use_pos_ngrams=base.use_pos_ngrams,
+            lsa_variants=base.lsa_variants,
+            lsa_components=base.lsa_components,
+            stylo_scaler=base.stylo_scaler,
+            stylo_normalize=base.stylo_normalize,
+            include_stylo=True,
+            include_lsa=base.include_lsa,
+            char_max_features=base.char_max_features,
+            pos_max_features=base.pos_max_features,
+            char_ngram_range=base.char_ngram_range,
+            pos_ngram_range=base.pos_ngram_range,
+            min_df=base.min_df,
+            random_state=base.random_state,
+        )
+        pipeline = build_pipeline(config, args.model)
+        result = run_repeated_cv(
+            pipeline, frame, y, n_splits=args.folds, n_repeats=args.repeats,
+            seed=args.seed, model_name=args.model, verbose=False,
+        )
+        pooled, _ = result.pooled_summary()
+        first = pooled.iloc[0].to_dict() if not pooled.empty else {}
+        rows.append({"variant": name, "macro_f1": first.get("macro_f1"), "balanced_accuracy": first.get("balanced_accuracy"), "accuracy": first.get("accuracy")})
+        print(f"[ablate] {name}: macro-F1={first.get('macro_f1', float('nan')):.4f}")
+    table = pd.DataFrame(rows)
+    full_value = table.loc[table["variant"] == "full", "macro_f1"].iloc[0]
+    table["delta_macro_f1"] = table["macro_f1"] - full_value
+    write_csv(table, output_dir / "ablation.csv")
+    print(f"[ablate] результаты: {output_dir / 'ablation.csv'}")
+    return 0
+
+
+def cmd_lsa_sweep(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    y = frame["character"].values
+    variants = args.variants.split(",")
+    components = [int(value) for value in args.components.split(",")]
+    output_dir = DEFAULT_RESULTS_DIR / _run_id("lsa_sweep")
+    output_dir.mkdir(parents=True, exist_ok=True)
+    rows = []
+    for variant in variants:
+        for k in components:
+            print(f"[lsa-sweep] {variant}, k={k}")
+            config = FeatureConfig(
+                groups=(),
+                use_char_ngrams=False,
+                use_pos_ngrams=False,
+                lsa_variants=(variant,),
+                lsa_components=k,
+                include_stylo=False,
+                include_lsa=True,
+            )
+            pipeline = build_pipeline(config, args.model)
+            result = run_repeated_cv(
+                pipeline, frame, y, n_splits=args.folds, n_repeats=args.repeats,
+                seed=args.seed, model_name=args.model, verbose=False,
+            )
+            pooled, _ = result.pooled_summary()
+            first = pooled.iloc[0].to_dict() if not pooled.empty else {}
+            rows.append({"variant": variant, "n_components": k, "macro_f1": first.get("macro_f1"), "accuracy": first.get("accuracy"), "top3_accuracy": first.get("top3_accuracy")})
+    table = pd.DataFrame(rows).sort_values(["macro_f1"], ascending=False)
+    write_csv(table, output_dir / "lsa_sweep.csv")
+    print(table.to_string(index=False))
+    return 0
+
+
+def cmd_delta(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    y = frame["character"].values
+    output_dir = DEFAULT_RESULTS_DIR / _run_id("delta")
+    output_dir.mkdir(parents=True, exist_ok=True)
+    rows = []
+    for n_features in [int(value) for value in args.features.split(",")]:
+        for metric in args.metrics.split(","):
+            print(f"[delta] MFW={n_features}, metric={metric}")
+            classifier = DeltaClassifier(n_features=n_features, metric=metric)
+            result = run_repeated_cv(
+                classifier, frame, y, n_splits=args.folds, n_repeats=args.repeats,
+                seed=args.seed, model_name=f"delta_{metric}", verbose=False,
+            )
+            pooled, _ = result.pooled_summary()
+            first = pooled.iloc[0].to_dict() if not pooled.empty else {}
+            rows.append({"n_features": n_features, "metric": metric, "macro_f1": first.get("macro_f1"), "accuracy": first.get("accuracy"), "balanced_accuracy": first.get("balanced_accuracy")})
+    table = pd.DataFrame(rows).sort_values("macro_f1", ascending=False)
+    write_csv(table, output_dir / "delta_sweep.csv")
+    print(table.to_string(index=False))
+    return 0
+
+
+def cmd_verify(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    y = frame["character"].values
+    feature_config = _feature_config(args.preset)
+    result = run_verification(frame, y, feature_config, n_splits=args.folds, seed=args.seed)
+    output_dir = DEFAULT_RESULTS_DIR / _run_id(f"verification_{args.preset}")
+    output_dir.mkdir(parents=True, exist_ok=True)
+    write_csv(result.fold_metrics, output_dir / "fold_metrics.csv")
+    result.scores.to_csv(output_dir / "pair_scores.csv", index=False, encoding="utf-8-sig")
+    save_json(result.summary, output_dir / "summary.json")
+    print(json.dumps(result.summary, ensure_ascii=False, indent=2))
+    return 0
+
+
+def cmd_cluster(args: argparse.Namespace) -> int:
+    frame = _frame(args)
+    y = frame["character"].values
+    feature_config = _feature_config(args.preset)
+    result = run_clustering(frame, y, feature_config, n_clusters=args.clusters, seed=args.seed)
+    output_dir = DEFAULT_RESULTS_DIR / _run_id(f"clustering_{args.preset}")
+    output_dir.mkdir(parents=True, exist_ok=True)
+    write_csv(result.metrics, output_dir / "clustering_metrics.csv")
+    result.centroids.to_csv(output_dir / "character_centroids.csv", encoding="utf-8-sig")
+    plot_pca(
+        result.centroids.to_numpy(),
+        result.centroids.index.tolist(),
+        output_dir / "figures" / "character_centroids_pca.png",
+        title="PCA центроидов персонажей",
+    )
+    print(result.metrics.to_string(index=False))
+    return 0
+
+
+def cmd_length_report(args: argparse.Namespace) -> int:
+    import pandas as pd
+
+    run_dir = Path(args.run_dir)
+    oof_path = run_dir / "oof_predictions.csv"
+    if not oof_path.exists():
+        raise FileNotFoundError(f"Нет файла предсказаний: {oof_path}")
+    oof = pd.read_csv(oof_path)
+    lines_path = Path(args.lines) if args.lines else PROJECT_ROOT / "data" / "processed" / "lines.csv"
+    if not lines_path.exists():
+        raise FileNotFoundError(
+            f"Нет файла {lines_path}. Сначала выполните python -m stylometry.cli audit."
+        )
+    lines = pd.read_csv(lines_path)
+    if "n_words" not in lines.columns:
+        lines["n_words"] = lines["text"].astype(str).str.findall(r"[A-Za-z0-9']+").map(len)
+    first = oof[oof["repeat"] == oof["repeat"].min()].copy()
+    first = first.merge(lines[["line_id", "n_words"]], on="line_id", how="left")
+    first["length_bin"] = pd.cut(
+        first["n_words"],
+        bins=[0, 5, 10, 15, 20, 30, 1000],
+        labels=["0–5", "6–10", "11–15", "16–20", "21–30", "31+"],
+        include_lowest=True,
+    )
+    labels = sorted(first["y_true"].unique())
+    rows = []
+    for bin_name, group in first.groupby("length_bin", observed=True):
+        # В узком диапазоне длины могут отсутствовать некоторые персонажи.
+        # Macro-F1 считаем только по классам, реально присутствующим в группе.
+        local_labels = sorted(group["y_true"].unique())
+        metrics = classification_metrics(
+            group["y_true"].values, group["y_pred"].values, labels=local_labels
+        )
+        interval = bootstrap_ci(
+            group["y_true"].values, group["y_pred"].values, local_labels,
+            metric="macro_f1", n_boot=500, seed=args.seed,
+        )
+        rows.append(
+            {
+                "length_bin": str(bin_name),
+                "n_samples": int(len(group)),
+                "macro_f1": metrics["macro_f1"],
+                "macro_f1_lower": interval["lower"],
+                "macro_f1_upper": interval["upper"],
+                "balanced_accuracy": metrics["balanced_accuracy"],
+                "accuracy": metrics["accuracy"],
+                "top3_accuracy": metrics.get("top3_accuracy"),
+            }
+        )
+    table = pd.DataFrame(rows)
+    output_path = run_dir / "length_bins.csv"
+    write_csv(table, output_path)
+    print(table.to_string(index=False))
+    print(f"[length-report] результат: {output_path}")
+    return 0
+
+
+def cmd_compare(args: argparse.Namespace) -> int:
+    import pandas as pd
+
+    def _read_first(path: str | Path) -> pd.DataFrame:
+        frame = pd.read_csv(Path(path) / "oof_predictions.csv")
+        return frame[frame["repeat"] == frame["repeat"].min()].copy()
+
+    first = _read_first(args.run_a).rename(columns={"y_pred": "pred_a"})
+    second = _read_first(args.run_b).rename(columns={"y_pred": "pred_b"})
+    merged = first.merge(
+        second[["line_id", "y_true", "pred_b"]],
+        on=["line_id", "y_true"],
+        how="inner",
+    )
+    if merged.empty:
+        raise ValueError("Не удалось сопоставить предсказания по line_id и y_true.")
+    labels = sorted(merged["y_true"].unique())
+    metrics_a = classification_metrics(merged["y_true"], merged["pred_a"], labels=labels)
+    metrics_b = classification_metrics(merged["y_true"], merged["pred_b"], labels=labels)
+    comparison = {
+        "n_matched": int(len(merged)),
+        "model_a": {
+            "macro_f1": metrics_a["macro_f1"],
+            "accuracy": metrics_a["accuracy"],
+            "balanced_accuracy": metrics_a["balanced_accuracy"],
+        },
+        "model_b": {
+            "macro_f1": metrics_b["macro_f1"],
+            "accuracy": metrics_b["accuracy"],
+            "balanced_accuracy": metrics_b["balanced_accuracy"],
+        },
+        "mcnemar": mcnemar_exact(merged["y_true"], merged["pred_a"], merged["pred_b"]),
+        "bootstrap_difference_macro_f1": bootstrap_difference(
+            merged["y_true"], merged["pred_a"], merged["pred_b"],
+            metric="macro_f1", labels=labels, n_boot=1000, seed=args.seed,
+        ),
+    }
+    output = {
+        "run_a": str(args.run_a),
+        "run_b": str(args.run_b),
+        "comparison": comparison,
+    }
+    path = Path(args.out) if args.out else Path(args.run_a) / "comparison_with_other.json"
+    save_json(output, path)
+    print(json.dumps(comparison, ensure_ascii=False, indent=2, default=_json_default))
+    print(f"[compare] результат: {path}")
+    return 0
+
+
+def cmd_subgroup_report(args: argparse.Namespace) -> int:
+    import pandas as pd
+
+    run_dir = Path(args.run_dir)
+    oof_path = run_dir / "oof_predictions.csv"
+    if not oof_path.exists():
+        raise FileNotFoundError(f"Нет файла предсказаний: {oof_path}")
+    oof = pd.read_csv(oof_path)
+    first = oof[oof["repeat"] == oof["repeat"].min()].copy()
+    labels = sorted(first["y_true"].unique())
+    metadata = load_character_metadata(args.metadata)
+    metrics = subgroup_report(
+        first["y_true"], first["y_pred"], labels, metadata, args.group
+    )
+    confusions = group_confusions(
+        first["y_true"], first["y_pred"], metadata, args.group
+    )
+    write_csv(metrics, run_dir / f"subgroup_{args.group}_metrics.csv")
+    write_csv(confusions, run_dir / f"subgroup_{args.group}_confusions.csv")
+    print(metrics.to_string(index=False))
+    print(f"[subgroup-report] результаты: {run_dir}")
+    return 0
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        description="Стилометрия реплик Stardew Valley: аудит, признаки, LSA и эксперименты."
+    )
+    subparsers = parser.add_subparsers(dest="command", required=True)
+
+    def add_common(sub: argparse.ArgumentParser) -> None:
+        sub.add_argument("--data", default=None, help="Путь к dialogues.csv или .xlsx")
+        sub.add_argument("--cache", default=None, help="Путь к кэшу токенизации")
+        sub.add_argument("--no-cache", action="store_true", help="Пересобрать кэш")
+        sub.add_argument("--out", default=None, help="Каталог для результатов")
+
+    audit = subparsers.add_parser("audit", help="Аудит корпуса")
+    add_common(audit)
+    audit.set_defaults(func=cmd_audit)
+
+    features = subparsers.add_parser("features", help="Ручные стилометрические признаки")
+    add_common(features)
+    features.set_defaults(func=cmd_features)
+
+    lsa = subparsers.add_parser("lsa", help="LSA одного варианта")
+    add_common(lsa)
+    lsa.add_argument("--variant", default="word", choices=LSA_VARIANTS)
+    lsa.add_argument("--components", type=int, default=100)
+    lsa.set_defaults(func=cmd_lsa)
+
+    run = subparsers.add_parser("run", help="Кросс-валидация одной модели")
+    add_common(run)
+    run.add_argument("--config", default=None)
+    run.add_argument("--preset", default=None)
+    run.add_argument("--model", default=None, choices=MODEL_NAMES)
+    run.add_argument("--folds", type=int, default=None)
+    run.add_argument("--repeats", type=int, default=None)
+    run.add_argument("--seed", type=int, default=None)
+    run.add_argument("--nested", action="store_true", help="Вложенная кросс-валидация")
+    run.set_defaults(func=cmd_run)
+
+    benchmark = subparsers.add_parser("benchmark", help="Сравнение классификаторов")
+    add_common(benchmark)
+    benchmark.add_argument("--preset", default="stylometry_pos_lsa")
+    benchmark.add_argument("--models", default=None, help="Список через запятую")
+    benchmark.add_argument("--folds", type=int, default=5)
+    benchmark.add_argument("--repeats", type=int, default=3)
+    benchmark.add_argument("--seed", type=int, default=42)
+    benchmark.set_defaults(func=cmd_benchmark)
+
+    ablate = subparsers.add_parser("ablate", help="Поочерёдное исключение групп признаков")
+    add_common(ablate)
+    ablate.add_argument("--preset", default="stylometry_pos_lsa")
+    ablate.add_argument("--model", default="logreg", choices=MODEL_NAMES)
+    ablate.add_argument("--folds", type=int, default=5)
+    ablate.add_argument("--repeats", type=int, default=2)
+    ablate.add_argument("--seed", type=int, default=42)
+    ablate.set_defaults(func=cmd_ablate)
+
+    sweep = subparsers.add_parser("lsa-sweep", help="Перебор вариантов и числа компонент LSA")
+    add_common(sweep)
+    sweep.add_argument("--variants", default="word,pos,char")
+    sweep.add_argument("--components", default="50,100,200")
+    sweep.add_argument("--model", default="logreg", choices=MODEL_NAMES)
+    sweep.add_argument("--folds", type=int, default=5)
+    sweep.add_argument("--repeats", type=int, default=1)
+    sweep.add_argument("--seed", type=int, default=42)
+    sweep.set_defaults(func=cmd_lsa_sweep)
+
+    delta = subparsers.add_parser("delta", help="Дельта Барроуза")
+    add_common(delta)
+    delta.add_argument("--features", default="100,200,500")
+    delta.add_argument("--metrics", default="manhattan,cosine")
+    delta.add_argument("--folds", type=int, default=5)
+    delta.add_argument("--repeats", type=int, default=2)
+    delta.add_argument("--seed", type=int, default=42)
+    delta.set_defaults(func=cmd_delta)
+
+    verify = subparsers.add_parser("verify", help="Верификация «свой/чужой»")
+    add_common(verify)
+    verify.add_argument("--preset", default="stylometry_pos_lsa")
+    verify.add_argument("--folds", type=int, default=5)
+    verify.add_argument("--seed", type=int, default=42)
+    verify.set_defaults(func=cmd_verify)
+
+    cluster = subparsers.add_parser("cluster", help="Кластеризация персонажей")
+    add_common(cluster)
+    cluster.add_argument("--preset", default="stylometry_pos_lsa")
+    cluster.add_argument("--clusters", type=int, default=None)
+    cluster.add_argument("--seed", type=int, default=42)
+    cluster.set_defaults(func=cmd_cluster)
+
+    length_report = subparsers.add_parser(
+        "length-report", help="Метрики по группам длины реплик"
+    )
+    length_report.add_argument("--run-dir", required=True)
+    length_report.add_argument("--lines", default=None)
+    length_report.add_argument("--seed", type=int, default=42)
+    length_report.set_defaults(func=cmd_length_report)
+
+    compare = subparsers.add_parser(
+        "compare", help="Парное сравнение двух запусков"
+    )
+    compare.add_argument("--run-a", required=True)
+    compare.add_argument("--run-b", required=True)
+    compare.add_argument("--out", default=None)
+    compare.add_argument("--seed", type=int, default=42)
+    compare.set_defaults(func=cmd_compare)
+
+    subgroup = subparsers.add_parser(
+        "subgroup-report", help="Метрики по полу, возрасту и виду персонажей"
+    )
+    subgroup.add_argument("--run-dir", required=True)
+    subgroup.add_argument("--metadata", default=None)
+    subgroup.add_argument(
+        "--group", default="gender", choices=["gender", "age_group", "species"]
+    )
+    subgroup.set_defaults(func=cmd_subgroup_report)
+
+    return parser
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+    args = parser.parse_args(argv)
+    return int(args.func(args))
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/lab1/stylometry/clustering.py b/lab1/stylometry/clustering.py
new file mode 100644
index 0000000..d577baf
--- /dev/null
+++ b/lab1/stylometry/clustering.py
@@ -0,0 +1,90 @@
+"""Кластеризация персонажей в стилометрическом пространстве."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, Sequence
+
+import numpy as np
+import pandas as pd
+from scipy.sparse import issparse
+from sklearn.cluster import AgglomerativeClustering, KMeans
+from sklearn.metrics import (
+    adjusted_rand_score,
+    calinski_harabasz_score,
+    davies_bouldin_score,
+    normalized_mutual_info_score,
+    silhouette_score,
+)
+from sklearn.preprocessing import normalize
+
+from .pipelines import FeatureConfig, build_feature_union
+
+
+@dataclass
+class ClusteringResult:
+    metrics: pd.DataFrame
+    centroids: pd.DataFrame
+
+
+def _centroid_matrix(matrix: np.ndarray, labels: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
+    classes = np.unique(labels)
+    centroids = np.vstack([matrix[labels == label].mean(axis=0) for label in classes])
+    return classes, centroids
+
+
+def run_clustering(
+    X: pd.DataFrame,
+    y: Sequence,
+    feature_config: FeatureConfig,
+    n_clusters: int | None = None,
+    seed: int = 42,
+) -> ClusteringResult:
+    """Кластеризует центроиды персонажей и считает внешние метрики."""
+    y = np.asarray(y)
+    features = build_feature_union(feature_config)
+    matrix = features.fit_transform(X)
+    matrix = matrix.toarray() if issparse(matrix) else np.asarray(matrix)
+    matrix = normalize(matrix, norm="l2", copy=False)
+    classes, centroids = _centroid_matrix(matrix, y)
+    # Если число кластеров равно числу персонажей, кластеризация тривиальна,
+    # а силуэт не определён. По умолчанию берём примерно корень из числа
+    # классов: для 34 персонажей это 6 групп.
+    n_clusters = int(n_clusters or max(2, round(len(classes) ** 0.5)))
+    n_clusters = min(n_clusters, len(classes) - 1)
+
+    rows: list[dict[str, Any]] = []
+    for name, model in (
+        (
+            "agglomerative_cosine",
+            AgglomerativeClustering(
+                n_clusters=n_clusters, metric="cosine", linkage="average"
+            ),
+        ),
+        (
+            "kmeans",
+            KMeans(n_clusters=n_clusters, random_state=seed, n_init=20),
+        ),
+    ):
+        predicted = model.fit_predict(centroids)
+        n_predicted = len(set(predicted))
+        valid_partition = 2 <= n_predicted < len(centroids)
+        rows.append(
+            {
+                "method": name,
+                "n_clusters": n_clusters,
+                "adjusted_rand_index": float(adjusted_rand_score(classes, predicted)),
+                "normalized_mutual_info": float(normalized_mutual_info_score(classes, predicted)),
+                "silhouette_cosine": float(
+                    silhouette_score(centroids, predicted, metric="cosine")
+                ) if valid_partition else 0.0,
+                "calinski_harabasz": float(calinski_harabasz_score(centroids, predicted))
+                if valid_partition else 0.0,
+                "davies_bouldin": float(davies_bouldin_score(centroids, predicted))
+                if valid_partition else 0.0,
+            }
+        )
+
+    centroid_frame = pd.DataFrame(centroids, index=classes)
+    centroid_frame.index.name = "character"
+    return ClusteringResult(metrics=pd.DataFrame(rows), centroids=centroid_frame)
diff --git a/lab1/stylometry/cv.py b/lab1/stylometry/cv.py
new file mode 100644
index 0000000..be96da5
--- /dev/null
+++ b/lab1/stylometry/cv.py
@@ -0,0 +1,293 @@
+"""Схемы кросс-валидации и сбор out-of-fold предсказаний."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any, Iterable, Sequence
+
+import numpy as np
+import pandas as pd
+from sklearn.base import clone
+from sklearn.model_selection import StratifiedKFold
+
+from .evaluation import (
+    align_scores,
+    classification_metrics,
+    metrics_summary,
+    summarize_pooled_predictions,
+)
+from .pipelines import make_grid_search
+
+
+@dataclass
+class CVResult:
+    """Результат повторной стратифицированной кросс-валидации."""
+
+    fold_metrics: pd.DataFrame
+    oof: pd.DataFrame
+    labels: np.ndarray
+    scores_by_repeat: dict[int, np.ndarray] = field(default_factory=dict)
+    n_splits: int = 5
+    n_repeats: int = 1
+    model_name: str = ""
+
+    def summary(self) -> pd.DataFrame:
+        return metrics_summary(self.fold_metrics)
+
+    def pooled_summary(self) -> tuple[pd.DataFrame, dict[str, float]]:
+        return summarize_pooled_predictions(
+            self.oof, self.labels, self.scores_by_repeat
+        )
+
+
+@dataclass
+class NestedCVResult(CVResult):
+    """Результат вложенной кросс-валидации с настройкой гиперпараметров."""
+
+    best_params: list[dict[str, Any]] = field(default_factory=list)
+
+
+def _predict_scores(pipeline: Any, X: pd.DataFrame) -> np.ndarray | None:
+    if hasattr(pipeline, "predict_proba"):
+        try:
+            return np.asarray(pipeline.predict_proba(X), dtype=float)
+        except Exception:
+            pass
+    if hasattr(pipeline, "decision_function"):
+        try:
+            scores = np.asarray(pipeline.decision_function(X), dtype=float)
+            if scores.ndim == 1:
+                scores = np.column_stack([-scores, scores])
+            return scores
+        except Exception:
+            pass
+    return None
+
+
+def _model_classes(pipeline: Any) -> np.ndarray:
+    if hasattr(pipeline, "named_steps") and "model" in pipeline.named_steps:
+        return np.asarray(pipeline.named_steps["model"].classes_)
+    return np.asarray(pipeline.classes_)
+
+
+def validate_folds(
+    y: Sequence,
+    splits: Iterable[tuple[np.ndarray, np.ndarray]],
+    labels: Sequence,
+) -> None:
+    """Проверяет, что каждый класс есть и в обучении, и в тесте."""
+    y = np.asarray(y)
+    labels_set = set(labels)
+    for fold_index, (train_idx, test_idx) in enumerate(splits):
+        train_labels = set(y[train_idx])
+        test_labels = set(y[test_idx])
+        missing_train = labels_set - train_labels
+        missing_test = labels_set - test_labels
+        if missing_train or missing_test:
+            raise ValueError(
+                f"Фолд {fold_index}: в обучении нет классов {sorted(missing_train)}, "
+                f"в тесте нет классов {sorted(missing_test)}. "
+                "Увеличьте размер выборки или уменьшите число фолдов."
+            )
+
+
+def _line_ids(X: pd.DataFrame, indices: np.ndarray) -> np.ndarray:
+    if "line_id" in X.columns:
+        return X.iloc[indices]["line_id"].to_numpy()
+    return np.asarray(indices)
+
+
+def _order_scores_like_oof(X: pd.DataFrame, rows: list[dict[str, Any]]) -> np.ndarray:
+    """Возвращает порядок строк матрицы score-ов, соответствующий OOF-таблице."""
+    if "line_id" in X.columns:
+        positions = {
+            int(value): index for index, value in enumerate(X["line_id"].to_numpy())
+        }
+        return np.asarray([positions[int(row["line_id"])] for row in rows], dtype=int)
+    return np.asarray([int(row["line_id"]) for row in rows], dtype=int)
+
+
+def run_repeated_cv(
+    pipeline: Any,
+    X: pd.DataFrame,
+    y: Sequence,
+    n_splits: int = 5,
+    n_repeats: int = 5,
+    seed: int = 42,
+    model_name: str = "",
+    verbose: bool = True,
+) -> CVResult:
+    """Повторная стратифицированная кросс-валидация.
+
+    Все преобразования внутри ``pipeline`` обучаются только на обучающем
+    блоке. Out-of-fold предсказания собираются для каждого повтора.
+    """
+    y = np.asarray(y)
+    labels = np.unique(y)
+    n_samples = len(y)
+    fold_rows: list[dict[str, Any]] = []
+    oof_rows: list[dict[str, Any]] = []
+    scores_by_repeat: dict[int, np.ndarray] = {}
+
+    for repeat in range(n_repeats):
+        splitter = StratifiedKFold(
+            n_splits=n_splits, shuffle=True, random_state=seed + repeat
+        )
+        splits = list(splitter.split(X, y))
+        validate_folds(y, splits, labels)
+        scores_oof = np.full((n_samples, len(labels)), np.nan, dtype=float)
+        oof_start = len(oof_rows)
+
+        for fold, (train_idx, test_idx) in enumerate(splits, start=1):
+            model = clone(pipeline)
+            model.fit(X.iloc[train_idx], y[train_idx])
+            y_pred = model.predict(X.iloc[test_idx])
+            raw_scores = _predict_scores(model, X.iloc[test_idx])
+            if raw_scores is not None:
+                scores = align_scores(
+                    raw_scores, _model_classes(model), labels, fill_value=0.0
+                )
+                scores_oof[test_idx] = scores
+            else:
+                scores = None
+
+            metrics = classification_metrics(
+                y[test_idx], y_pred, y_score=scores, labels=labels
+            )
+            metrics.update(
+                {
+                    "repeat": repeat,
+                    "fold": fold,
+                    "n_train": int(len(train_idx)),
+                    "n_test": int(len(test_idx)),
+                }
+            )
+            fold_rows.append(metrics)
+
+            for row_index, sample_index in enumerate(test_idx):
+                oof_rows.append(
+                    {
+                        "repeat": repeat,
+                        "fold": fold,
+                        "line_id": int(_line_ids(X, np.asarray([sample_index]))[0]),
+                        "character": str(X.iloc[sample_index].get("character", "")),
+                        "y_true": y[sample_index],
+                        "y_pred": y_pred[row_index],
+                    }
+                )
+            if verbose:
+                print(
+                    f"[cv] повтор {repeat + 1}/{n_repeats}, "
+                    f"фолд {fold}/{n_splits}: macro-F1={metrics['macro_f1']:.4f}"
+                )
+
+        if not np.isnan(scores_oof).any():
+            repeat_rows = sorted(oof_rows[oof_start:], key=lambda row: int(row["line_id"]))
+            order = _order_scores_like_oof(X, repeat_rows)
+            scores_by_repeat[repeat] = scores_oof[order]
+
+    oof = pd.DataFrame(oof_rows)
+    if not oof.empty:
+        oof = oof.sort_values(["repeat", "line_id"]).reset_index(drop=True)
+    return CVResult(
+        fold_metrics=pd.DataFrame(fold_rows),
+        oof=oof,
+        labels=labels,
+        scores_by_repeat=scores_by_repeat,
+        n_splits=n_splits,
+        n_repeats=n_repeats,
+        model_name=model_name,
+    )
+
+
+def run_nested_cv(
+    pipeline: Any,
+    X: pd.DataFrame,
+    y: Sequence,
+    model_name: str,
+    outer_splits: int = 5,
+    inner_splits: int = 3,
+    seed: int = 42,
+    verbose: bool = True,
+) -> NestedCVResult:
+    """Вложенная кросс-валидация: настройка гиперпараметров внутри outer-фолдов."""
+    y = np.asarray(y)
+    labels = np.unique(y)
+    n_samples = len(y)
+    outer = StratifiedKFold(
+        n_splits=outer_splits, shuffle=True, random_state=seed
+    )
+    splits = list(outer.split(X, y))
+    validate_folds(y, splits, labels)
+
+    fold_rows: list[dict[str, Any]] = []
+    oof_rows: list[dict[str, Any]] = []
+    best_params: list[dict[str, Any]] = []
+    scores_oof = np.full((n_samples, len(labels)), np.nan, dtype=float)
+
+    for fold, (train_idx, test_idx) in enumerate(splits, start=1):
+        search = make_grid_search(
+            pipeline, model_name, cv=inner_splits, scoring="f1_macro", n_jobs=-1
+        )
+        search.fit(X.iloc[train_idx], y[train_idx])
+        model = search.best_estimator_
+        y_pred = model.predict(X.iloc[test_idx])
+        raw_scores = _predict_scores(model, X.iloc[test_idx])
+        if raw_scores is not None:
+            scores = align_scores(
+                raw_scores, _model_classes(model), labels, fill_value=0.0
+            )
+            scores_oof[test_idx] = scores
+        else:
+            scores = None
+
+        metrics = classification_metrics(
+            y[test_idx], y_pred, y_score=scores, labels=labels
+        )
+        metrics.update(
+            {
+                "repeat": 0,
+                "fold": fold,
+                "n_train": int(len(train_idx)),
+                "n_test": int(len(test_idx)),
+            }
+        )
+        fold_rows.append(metrics)
+        best_params.append(dict(search.best_params_))
+        for row_index, sample_index in enumerate(test_idx):
+            oof_rows.append(
+                {
+                    "repeat": 0,
+                    "fold": fold,
+                    "line_id": int(_line_ids(X, np.asarray([sample_index]))[0]),
+                    "character": str(X.iloc[sample_index].get("character", "")),
+                    "y_true": y[sample_index],
+                    "y_pred": y_pred[row_index],
+                }
+            )
+        if verbose:
+            print(
+                f"[nested-cv] фолд {fold}/{outer_splits}: "
+                f"macro-F1={metrics['macro_f1']:.4f}, "
+                f"best={search.best_params_}"
+            )
+
+    scores_by_repeat = {}
+    if not np.isnan(scores_oof).any():
+        ordered_rows = sorted(oof_rows, key=lambda row: int(row["line_id"]))
+        order = _order_scores_like_oof(X, ordered_rows)
+        scores_by_repeat[0] = scores_oof[order]
+
+    oof = pd.DataFrame(oof_rows)
+    if not oof.empty:
+        oof = oof.sort_values(["repeat", "line_id"]).reset_index(drop=True)
+    return NestedCVResult(
+        fold_metrics=pd.DataFrame(fold_rows),
+        oof=oof,
+        labels=labels,
+        scores_by_repeat=scores_by_repeat,
+        n_splits=outer_splits,
+        n_repeats=1,
+        model_name=model_name,
+        best_params=best_params,
+    )
diff --git a/lab1/stylometry/datasets.py b/lab1/stylometry/datasets.py
new file mode 100644
index 0000000..cc9b93d
--- /dev/null
+++ b/lab1/stylometry/datasets.py
@@ -0,0 +1,86 @@
+"""Подготовка таблицы реплик с токенами, тегами и текстами для LSA."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pandas as pd
+
+from .io import DEFAULT_DATA_PATH, corpus_fingerprint, read_corpus
+from .tokenization import Tokenizer
+
+DEFAULT_CACHE_PATH = Path(__file__).resolve().parents[1] / "data" / "processed" / "tokenized.pkl"
+
+REQUIRED_COLUMNS = {
+    "line_id",
+    "character",
+    "text",
+    "tokens",
+    "ptb",
+    "upos",
+    "sentences",
+    "pos_text",
+    "fw_text",
+    "content_text",
+}
+
+
+def build_feature_frame(
+    data_path: str | Path | None = None,
+    tokenizer: Tokenizer | None = None,
+    verbose: bool = True,
+) -> pd.DataFrame:
+    """Читает корпус и добавляет результаты токенизации и разметки."""
+    data_path = Path(data_path) if data_path is not None else DEFAULT_DATA_PATH
+    frame = read_corpus(data_path)
+    tokenizer = tokenizer or Tokenizer(use_nltk=True)
+    features = tokenizer.analyze_frame(frame, verbose=verbose)
+    features.attrs["corpus_path"] = str(data_path)
+    features.attrs["corpus_sha256"] = corpus_fingerprint(data_path)
+    features.attrs["nltk_used"] = tokenizer.use_nltk
+    features.attrs["pos_available"] = tokenizer.pos_available
+    return features
+
+
+def save_feature_frame(frame: pd.DataFrame, path: str | Path) -> Path:
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    frame.to_pickle(path)
+    return path
+
+
+def load_feature_frame(
+    data_path: str | Path | None = None,
+    cache_path: str | Path | None = None,
+    force: bool = False,
+    verbose: bool = True,
+) -> pd.DataFrame:
+    """Загружает кэш токенизации или строит его заново.
+
+    Кэш проверяется по SHA-256 исходного файла. Если файл корпуса изменился,
+    токенизация выполняется повторно.
+    """
+    data_path = Path(data_path) if data_path is not None else DEFAULT_DATA_PATH
+    cache_path = Path(cache_path) if cache_path is not None else DEFAULT_CACHE_PATH
+
+    if cache_path.exists() and not force:
+        try:
+            frame = pd.read_pickle(cache_path)
+            if REQUIRED_COLUMNS <= set(frame.columns):
+                cached_hash = frame.attrs.get("corpus_sha256")
+                current_hash = corpus_fingerprint(data_path)
+                if cached_hash == current_hash:
+                    if verbose:
+                        print(f"[datasets] загружен кэш: {cache_path}")
+                    return frame
+                if verbose:
+                    print("[datasets] корпус изменился, пересобираю признаки.")
+        except Exception as error:
+            if verbose:
+                print(f"[datasets] не удалось прочитать кэш ({error}), пересобираю.")
+
+    frame = build_feature_frame(data_path, verbose=verbose)
+    save_feature_frame(frame, cache_path)
+    if verbose:
+        print(f"[datasets] кэш сохранён: {cache_path}")
+    return frame
diff --git a/lab1/stylometry/delta.py b/lab1/stylometry/delta.py
new file mode 100644
index 0000000..66dbe6e
--- /dev/null
+++ b/lab1/stylometry/delta.py
@@ -0,0 +1,148 @@
+"""Классическая дельта Барроуза и её косинусный вариант.
+
+Дельта Барроуза — стандартный метод стилометрии. Он использует частоты
+наиболее частотных слов, z-нормирование по обучающей выборке и расстояние до
+центроида класса. Мы реализуем два варианта: манхэттенское расстояние
+(классическая дельта) и косинусное расстояние (Cosine Delta).
+"""
+
+from __future__ import annotations
+
+import re
+from collections import Counter
+from typing import Any, Sequence
+
+import numpy as np
+import pandas as pd
+from scipy.special import softmax
+from scipy.spatial.distance import cdist
+from sklearn.base import BaseEstimator, ClassifierMixin, TransformerMixin
+
+_TOKEN_RE = re.compile(r"[A-Za-z]+(?:'[A-Za-z]+)?")
+
+
+def _texts(X: Any) -> list[str]:
+    if isinstance(X, pd.DataFrame):
+        if "text" not in X.columns:
+            raise KeyError("Для Delta нужна колонка 'text'.")
+        return X["text"].astype(str).tolist()
+    return [str(value) for value in X]
+
+
+def _tokens(text: str, lowercase: bool = True) -> list[str]:
+    tokens = _TOKEN_RE.findall(text)
+    return [token.lower() for token in tokens] if lowercase else tokens
+
+
+class MFWVectorizer(BaseEstimator, TransformerMixin):
+    """Векторизатор по наиболее частотным словам (Most Frequent Words)."""
+
+    def __init__(
+        self,
+        n_features: int = 100,
+        lowercase: bool = True,
+        zscore: bool = True,
+    ) -> None:
+        self.n_features = n_features
+        self.lowercase = lowercase
+        self.zscore = zscore
+
+    def fit(self, X: Any, y=None) -> "MFWVectorizer":
+        texts = _texts(X)
+        counter: Counter[str] = Counter()
+        for text in texts:
+            counter.update(_tokens(text, self.lowercase))
+        self.vocabulary_ = [word for word, _ in counter.most_common(self.n_features)]
+        if not self.vocabulary_:
+            raise ValueError("Не удалось выделить слова для Delta.")
+        index = {word: position for position, word in enumerate(self.vocabulary_)}
+        raw = np.zeros((len(texts), len(self.vocabulary_)), dtype=float)
+        for row, text in enumerate(texts):
+            tokens = _tokens(text, self.lowercase)
+            counts = Counter(tokens)
+            total = max(len(tokens), 1)
+            for word, count in counts.items():
+                position = index.get(word)
+                if position is not None:
+                    raw[row, position] = count / total
+        self.mean_ = raw.mean(axis=0)
+        std = raw.std(axis=0, ddof=0)
+        self.std_ = np.where(std == 0, 1.0, std)
+        return self
+
+    def transform(self, X: Any) -> np.ndarray:
+        texts = _texts(X)
+        index = {word: position for position, word in enumerate(self.vocabulary_)}
+        raw = np.zeros((len(texts), len(self.vocabulary_)), dtype=float)
+        for row, text in enumerate(texts):
+            tokens = _tokens(text, self.lowercase)
+            counts = Counter(tokens)
+            total = max(len(tokens), 1)
+            for word, count in counts.items():
+                position = index.get(word)
+                if position is not None:
+                    raw[row, position] = count / total
+        if self.zscore:
+            return (raw - self.mean_) / self.std_
+        return raw
+
+    def get_feature_names_out(self, input_features=None) -> np.ndarray:
+        return np.asarray(self.vocabulary_, dtype=object)
+
+
+class DeltaClassifier(BaseEstimator, ClassifierMixin):
+    """Классификатор на основе расстояния до центроида класса."""
+
+    def __init__(
+        self,
+        n_features: int = 100,
+        metric: str = "manhattan",
+        zscore: bool = True,
+        lowercase: bool = True,
+    ) -> None:
+        self.n_features = n_features
+        self.metric = metric
+        self.zscore = zscore
+        self.lowercase = lowercase
+
+    def _make_vectorizer(self) -> MFWVectorizer:
+        return MFWVectorizer(
+            n_features=self.n_features,
+            lowercase=self.lowercase,
+            zscore=self.zscore,
+        )
+
+    def fit(self, X: Any, y: Sequence) -> "DeltaClassifier":
+        y = np.asarray(y)
+        self.vectorizer_ = self._make_vectorizer()
+        matrix = self.vectorizer_.fit_transform(X)
+        self.classes_ = np.unique(y)
+        self.centroids_ = np.vstack(
+            [matrix[y == label].mean(axis=0) for label in self.classes_]
+        )
+        return self
+
+    def _distances(self, X: Any) -> np.ndarray:
+        matrix = self.vectorizer_.transform(X)
+        if self.metric == "manhattan":
+            return cdist(matrix, self.centroids_, metric="cityblock") / max(matrix.shape[1], 1)
+        if self.metric == "euclidean":
+            return cdist(matrix, self.centroids_, metric="euclidean")
+        if self.metric == "cosine":
+            return cdist(matrix, self.centroids_, metric="cosine")
+        raise ValueError(f"Неизвестная метрика Delta: {self.metric}")
+
+    def decision_function(self, X: Any) -> np.ndarray:
+        return -self._distances(X)
+
+    def predict(self, X: Any) -> np.ndarray:
+        distances = self._distances(X)
+        return self.classes_[np.argmin(distances, axis=1)]
+
+    def predict_proba(self, X: Any) -> np.ndarray:
+        distances = self._distances(X)
+        scale = np.maximum(distances.std(axis=1, keepdims=True), 1e-6)
+        return softmax(-distances / scale, axis=1)
+
+    def get_feature_names_out(self) -> np.ndarray:
+        return self.vectorizer_.get_feature_names_out()
diff --git a/lab1/stylometry/evaluation.py b/lab1/stylometry/evaluation.py
new file mode 100644
index 0000000..89e6a1e
--- /dev/null
+++ b/lab1/stylometry/evaluation.py
@@ -0,0 +1,273 @@
+"""Метрики качества классификации и их агрегация."""
+
+from __future__ import annotations
+
+from typing import Iterable, Sequence
+
+import numpy as np
+import pandas as pd
+from sklearn.metrics import (
+    accuracy_score,
+    balanced_accuracy_score,
+    cohen_kappa_score,
+    confusion_matrix,
+    f1_score,
+    log_loss,
+    matthews_corrcoef,
+    precision_recall_fscore_support,
+    top_k_accuracy_score,
+)
+
+
+def _labels_or_default(y_true: Sequence, y_pred: Sequence, labels: Sequence | None) -> np.ndarray:
+    if labels is not None:
+        return np.asarray(labels)
+    return np.unique(np.concatenate([np.asarray(y_true), np.asarray(y_pred)]))
+
+
+def brier_score_multiclass(y_true: Sequence, y_score: np.ndarray, labels: Sequence) -> float:
+    """Многоклассовая оценка Брайера: средний квадрат отклонения вероятностей."""
+    label_to_index = {label: index for index, label in enumerate(labels)}
+    y_true = np.asarray(y_true)
+    one_hot = np.zeros_like(y_score, dtype=float)
+    for row, label in enumerate(y_true):
+        one_hot[row, label_to_index[label]] = 1.0
+    return float(np.mean(np.sum((y_score - one_hot) ** 2, axis=1)))
+
+
+def align_scores(
+    y_score: np.ndarray,
+    model_classes: Sequence,
+    labels: Sequence,
+    fill_value: float = 0.0,
+) -> np.ndarray:
+    """Приводит столбцы вероятностей/решений к единому порядку классов."""
+    labels = list(labels)
+    model_classes = list(model_classes)
+    result = np.full((y_score.shape[0], len(labels)), fill_value, dtype=float)
+    for source_index, label in enumerate(model_classes):
+        if label in labels:
+            result[:, labels.index(label)] = y_score[:, source_index]
+    return result
+
+
+def classification_metrics(
+    y_true: Sequence,
+    y_pred: Sequence,
+    y_score: np.ndarray | None = None,
+    labels: Sequence | None = None,
+    top_k: Iterable[int] = (3, 5),
+) -> dict[str, float]:
+    """Считает набор метрик для одного разбиения."""
+    y_true = np.asarray(y_true)
+    y_pred = np.asarray(y_pred)
+    labels_array = _labels_or_default(y_true, y_pred, labels)
+    n_classes = len(labels_array)
+
+    result: dict[str, float] = {
+        "n_samples": int(len(y_true)),
+        "n_classes": int(n_classes),
+        "accuracy": float(accuracy_score(y_true, y_pred)),
+        "balanced_accuracy": float(
+            balanced_accuracy_score(y_true, y_pred)
+        ),
+        "macro_f1": float(
+            f1_score(y_true, y_pred, labels=labels_array, average="macro", zero_division=0)
+        ),
+        "weighted_f1": float(
+            f1_score(y_true, y_pred, labels=labels_array, average="weighted", zero_division=0)
+        ),
+        "micro_f1": float(
+            f1_score(y_true, y_pred, labels=labels_array, average="micro", zero_division=0)
+        ),
+        "mcc": float(matthews_corrcoef(y_true, y_pred)),
+        "cohen_kappa": float(cohen_kappa_score(y_true, y_pred, labels=labels_array)),
+    }
+
+    if y_score is not None:
+        y_score = np.asarray(y_score, dtype=float)
+        if y_score.ndim == 2 and y_score.shape[1] == n_classes:
+            for k in top_k:
+                if 1 < k < n_classes:
+                    result[f"top{k}_accuracy"] = float(
+                        top_k_accuracy_score(y_true, y_score, k=k, labels=labels_array)
+                    )
+            row_sums = y_score.sum(axis=1)
+            if np.all(y_score >= 0) and np.allclose(row_sums, 1.0, atol=1e-5):
+                result["log_loss"] = float(log_loss(y_true, y_score, labels=labels_array))
+                result["brier_score"] = brier_score_multiclass(y_true, y_score, labels_array)
+
+    return result
+
+
+def per_class_report(
+    y_true: Sequence,
+    y_pred: Sequence,
+    labels: Sequence,
+) -> pd.DataFrame:
+    """Точность, полнота и F-мера по каждому классу."""
+    precision, recall, f1, support = precision_recall_fscore_support(
+        y_true, y_pred, labels=labels, zero_division=0
+    )
+    return pd.DataFrame(
+        {
+            "class": list(labels),
+            "precision": precision,
+            "recall": recall,
+            "f1": f1,
+            "support": support,
+        }
+    )
+
+
+def confusion_matrix_frame(
+    y_true: Sequence,
+    y_pred: Sequence,
+    labels: Sequence,
+    normalize: str | None = None,
+) -> pd.DataFrame:
+    """Матрица ошибок в виде таблицы с подписями классов."""
+    matrix = confusion_matrix(y_true, y_pred, labels=labels, normalize=normalize)
+    return pd.DataFrame(matrix, index=list(labels), columns=list(labels))
+
+
+def top_confusions(
+    y_true: Sequence,
+    y_pred: Sequence,
+    labels: Sequence,
+    top_n: int = 30,
+) -> pd.DataFrame:
+    """Самые частые пары «истинный класс — предсказанный класс»."""
+    frame = confusion_matrix_frame(y_true, y_pred, labels)
+    rows = []
+    for true_label in frame.index:
+        for predicted_label in frame.columns:
+            if true_label == predicted_label:
+                continue
+            count = int(frame.loc[true_label, predicted_label])
+            if count > 0:
+                rows.append(
+                    {
+                        "true_class": true_label,
+                        "predicted_class": predicted_label,
+                        "count": count,
+                        "share_of_true_class": count / max(int(frame.loc[true_label].sum()), 1),
+                    }
+                )
+    result = pd.DataFrame(rows)
+    if result.empty:
+        return result
+    return result.sort_values(["count", "share_of_true_class"], ascending=False).head(top_n)
+
+
+def bootstrap_ci(
+    y_true: Sequence,
+    y_pred: Sequence,
+    labels: Sequence,
+    metric: str = "macro_f1",
+    n_boot: int = 1000,
+    seed: int = 42,
+) -> dict[str, float]:
+    """Перцентильный доверительный интервал метрики по репликам."""
+    y_true = np.asarray(y_true)
+    y_pred = np.asarray(y_pred)
+    labels_array = np.asarray(labels)
+    rng = np.random.default_rng(seed)
+
+    def _metric(true_values: np.ndarray, predicted_values: np.ndarray) -> float:
+        if metric == "macro_f1":
+            return float(
+                f1_score(
+                    true_values,
+                    predicted_values,
+                    labels=labels_array,
+                    average="macro",
+                    zero_division=0,
+                )
+            )
+        if metric == "accuracy":
+            return float(accuracy_score(true_values, predicted_values))
+        if metric == "balanced_accuracy":
+            return float(balanced_accuracy_score(true_values, predicted_values))
+        if metric == "mcc":
+            return float(matthews_corrcoef(true_values, predicted_values))
+        raise ValueError(f"Неизвестная метрика для bootstrap: {metric}")
+
+    values = []
+    n_samples = len(y_true)
+    for _ in range(n_boot):
+        indices = rng.integers(0, n_samples, n_samples)
+        values.append(_metric(y_true[indices], y_pred[indices]))
+    values_array = np.asarray(values, dtype=float)
+    return {
+        "mean": float(np.mean(values_array)),
+        "lower": float(np.percentile(values_array, 2.5)),
+        "upper": float(np.percentile(values_array, 97.5)),
+        "n_boot": int(n_boot),
+    }
+
+
+def metrics_summary(
+    metrics: pd.DataFrame,
+    columns: Iterable[str] | None = None,
+) -> pd.DataFrame:
+    """Среднее и стандартное отклонение метрик по фолдам/повторам."""
+    if columns is None:
+        columns = [
+            column
+            for column in metrics.columns
+            if column not in {"repeat", "fold", "n_train", "n_test", "n_samples", "n_classes"}
+        ]
+    rows = []
+    for column in columns:
+        if column not in metrics.columns:
+            continue
+        values = pd.to_numeric(metrics[column], errors="coerce").dropna()
+        if values.empty:
+            continue
+        rows.append(
+            {
+                "metric": column,
+                "mean": float(values.mean()),
+                "std": float(values.std(ddof=1)) if len(values) > 1 else 0.0,
+                "min": float(values.min()),
+                "max": float(values.max()),
+                "n": int(len(values)),
+            }
+        )
+    return pd.DataFrame(rows)
+
+
+def summarize_pooled_predictions(
+    oof: pd.DataFrame,
+    labels: Sequence,
+    scores_by_repeat: dict[int, np.ndarray] | None = None,
+    top_k: Iterable[int] = (3, 5),
+) -> tuple[pd.DataFrame, dict[str, float]]:
+    """Считает метрики по pooled out-of-fold предсказаниям.
+
+    Возвращает таблицу метрик по повторам и итоговую bootstrap-оценку по
+    первому повтору (для доверительного интервала).
+    """
+    rows = []
+    for repeat, group in oof.groupby("repeat"):
+        scores = None
+        if scores_by_repeat is not None and repeat in scores_by_repeat:
+            scores = scores_by_repeat[repeat]
+        metrics = classification_metrics(
+            group["y_true"].values,
+            group["y_pred"].values,
+            y_score=scores,
+            labels=labels,
+            top_k=top_k,
+        )
+        metrics["repeat"] = int(repeat)
+        rows.append(metrics)
+    per_repeat = pd.DataFrame(rows)
+
+    first_repeat = oof[oof["repeat"] == oof["repeat"].min()]
+    bootstrap = {
+        metric: bootstrap_ci(first_repeat["y_true"].values, first_repeat["y_pred"].values, labels, metric)
+        for metric in ("macro_f1", "accuracy", "balanced_accuracy", "mcc")
+    }
+    return per_repeat, bootstrap
diff --git a/lab1/stylometry/io.py b/lab1/stylometry/io.py
new file mode 100644
index 0000000..991b3f4
--- /dev/null
+++ b/lab1/stylometry/io.py
@@ -0,0 +1,183 @@
+"""Чтение и валидация корпуса реплик Stardew Valley."""
+
+from __future__ import annotations
+
+import hashlib
+from pathlib import Path
+from typing import Any
+
+import pandas as pd
+
+from .cleaning import clean_text, count_placeholders, is_valid_line
+
+# Русские названия колонок в исходном CSV/XLSX.
+COLUMN_RENAMES = {
+    "Реплика": "text",
+    "Персонаж": "character",
+    "Text": "text",
+    "Character": "character",
+}
+
+DEFAULT_DATA_PATH = Path(__file__).resolve().parents[1] / "data" / "dialogues.csv"
+DEFAULT_STATS_PATH = Path(__file__).resolve().parents[1] / "data" / "stats_per_character.xlsx"
+
+
+class CorpusError(ValueError):
+    """Ошибка структуры или кодировки корпуса."""
+
+
+def _read_csv_robust(path: Path) -> pd.DataFrame:
+    """Пробует несколько кодировок и проверяет наличие нужных колонок.
+
+    Основная кодировка корпуса — cp1251. UTF-8 добавлен на случай, если файл
+    будет пересохранён; koi8-r — как редкий запасной вариант.
+    """
+    last_error: Exception | None = None
+    for encoding in ("cp1251", "utf-8-sig", "utf-8", "koi8-r"):
+        try:
+            frame = pd.read_csv(path, sep=";", encoding=encoding, dtype=str)
+        except (UnicodeDecodeError, UnicodeError) as error:
+            last_error = error
+            continue
+        except pd.errors.ParserError as error:
+            last_error = error
+            continue
+
+        columns = set(frame.columns)
+        if {"Реплика", "Персонаж"} <= columns or {"text", "character"} <= columns:
+            return frame
+
+        # Файл прочитался, но это не тот формат. Пробуем следующую кодировку.
+        last_error = CorpusError(
+            f"В файле {path} не найдены колонки 'Реплика'/'Персонаж' "
+            f"или 'text'/'character'. Найдены: {list(frame.columns)}"
+        )
+    raise CorpusError(f"Не удалось прочитать CSV {path}: {last_error}")
+
+
+def _read_excel(path: Path) -> pd.DataFrame:
+    """Читает первый лист Excel. Второй лист в исходном файле пустой."""
+    try:
+        return pd.read_excel(path, sheet_name=0, dtype=str)
+    except Exception as error:  # openpyxl поднимает разные типы исключений
+        raise CorpusError(f"Не удалось прочитать Excel {path}: {error}") from error
+
+
+def read_corpus(path: str | Path | None = None) -> pd.DataFrame:
+    """Читает корпус и возвращает таблицу с колонками ``text`` и ``character``.
+
+    Дополнительно сохраняются:
+    - ``text_raw`` — исходная реплика после приведения к строке;
+    - ``line_id`` — сквозной номер реплики;
+    - ``n_chars`` — длина нормализованного текста;
+    - ``placeholder_count`` — число плейсхолдеров.
+    """
+    path = Path(path) if path is not None else DEFAULT_DATA_PATH
+    if not path.exists():
+        raise FileNotFoundError(f"Файл корпуса не найден: {path}")
+
+    suffix = path.suffix.lower()
+    if suffix == ".csv":
+        frame = _read_csv_robust(path)
+    elif suffix in {".xlsx", ".xls"}:
+        frame = _read_excel(path)
+    else:
+        raise CorpusError(f"Неподдерживаемый формат файла: {suffix}")
+
+    frame = frame.rename(columns=COLUMN_RENAMES)
+    required = {"text", "character"}
+    if not required <= set(frame.columns):
+        raise CorpusError(
+            f"Ожидались колонки {sorted(required)}, найдены {list(frame.columns)}"
+        )
+
+    frame = frame[["text", "character"]].copy()
+    frame = frame.dropna(subset=["text", "character"])
+    frame["text"] = frame["text"].astype(str).str.strip()
+    frame["character"] = frame["character"].astype(str).str.strip()
+    frame = frame[(frame["text"] != "") & (frame["character"] != "")]
+
+    frame["text_raw"] = frame["text"]
+    frame["text"] = frame["text_raw"].map(clean_text)
+    frame = frame[frame["text"].map(is_valid_line)].copy()
+
+    frame = frame.reset_index(drop=True)
+    frame.insert(0, "line_id", range(len(frame)))
+    frame["n_chars"] = frame["text"].str.len().astype(int)
+    frame["n_words"] = frame["text"].str.findall(r"[A-Za-z0-9']+").map(len).astype(int)
+    frame["placeholder_count"] = frame["text_raw"].map(count_placeholders).astype(int)
+
+    if frame.empty:
+        raise CorpusError("После нормализации корпус оказался пустым.")
+
+    return frame
+
+
+def read_stats(path: str | Path | None = None) -> pd.DataFrame:
+    """Читает таблицу статистики по персонажам, если она есть."""
+    path = Path(path) if path is not None else DEFAULT_STATS_PATH
+    if not path.exists():
+        return pd.DataFrame()
+    return pd.read_excel(path, sheet_name=0)
+
+
+def validate_corpus(frame: pd.DataFrame) -> dict[str, Any]:
+    """Считает основные характеристики корпуса для отчёта и проверок."""
+    counts = frame["character"].value_counts().sort_values(ascending=False)
+    word_counts = frame["text"].str.findall(r"[A-Za-z0-9']+").map(len)
+    char_counts = frame["text"].str.len()
+
+    return {
+        "n_lines": int(len(frame)),
+        "n_characters": int(frame["character"].nunique()),
+        "n_words": int(word_counts.sum()),
+        "mean_words_per_line": float(word_counts.mean()),
+        "median_words_per_line": float(word_counts.median()),
+        "min_words_per_line": int(word_counts.min()),
+        "max_words_per_line": int(word_counts.max()),
+        "mean_chars_per_line": float(char_counts.mean()),
+        "median_chars_per_line": float(char_counts.median()),
+        "class_counts": counts.to_dict(),
+        "n_duplicate_texts": int(frame["text"].duplicated().sum()),
+        "n_placeholder_lines": int((frame["placeholder_count"] > 0).sum()),
+        "n_placeholders": int(frame["placeholder_count"].sum()),
+    }
+
+
+def compare_with_stats(frame: pd.DataFrame, stats: pd.DataFrame) -> pd.DataFrame:
+    """Сравнивает фактические counts с таблицей stats_per_character.xlsx.
+
+    Возвращает таблицу с колонками ``character``, ``actual``, ``reported``,
+    ``difference``. Если таблица статистики пуста, возвращается пустая таблица.
+    """
+    if stats.empty or "Персонаж" not in stats.columns:
+        return pd.DataFrame()
+
+    reported_column = "реплик" if "реплик" in stats.columns else stats.columns[1]
+    reported = (
+        stats[["Персонаж", reported_column]]
+        .rename(columns={"Персонаж": "character", reported_column: "reported"})
+        .set_index("character")["reported"]
+    )
+    actual = frame["character"].value_counts()
+    comparison = pd.DataFrame({"actual": actual, "reported": reported})
+    comparison["difference"] = comparison["actual"] - comparison["reported"]
+    return comparison.fillna(0).astype({"actual": int, "reported": int, "difference": int})
+
+
+def corpus_fingerprint(path: str | Path | None = None) -> str:
+    """SHA-256 файла корпуса для журнала воспроизводимости."""
+    path = Path(path) if path is not None else DEFAULT_DATA_PATH
+    digest = hashlib.sha256()
+    with path.open("rb") as stream:
+        for chunk in iter(lambda: stream.read(1024 * 1024), b""):
+            digest.update(chunk)
+    return digest.hexdigest()
+
+
+def write_csv(frame: pd.DataFrame, path: str | Path) -> Path:
+    """Сохраняет таблицу в UTF-8 с BOM, чтобы её открывал Excel."""
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    frame.to_csv(path, index=False, encoding="utf-8-sig")
+    return path
diff --git a/lab1/stylometry/lexicons.py b/lab1/stylometry/lexicons.py
new file mode 100644
index 0000000..7a5d3e4
--- /dev/null
+++ b/lab1/stylometry/lexicons.py
@@ -0,0 +1,175 @@
+"""Закрытые классы слов и небольшие лексиконы для стилометрии.
+
+Списки фиксированы заранее и не зависят от корпуса. Это важно для
+воспроизводимости: признаки не подбираются по тестовой выборке.
+"""
+
+from __future__ import annotations
+
+ARTICLES = {"a", "an", "the"}
+
+DETERMINERS = {
+    "this", "that", "these", "those",
+    "my", "your", "his", "her", "its", "our", "their",
+    "some", "any", "no", "every", "each", "either", "neither",
+    "both", "all", "few", "many", "much", "more", "most",
+    "other", "another", "such", "what", "which", "whose",
+}
+
+PREPOSITIONS = {
+    "about", "above", "across", "after", "against", "along", "among",
+    "around", "at", "before", "behind", "below", "beneath", "beside",
+    "between", "beyond", "by", "despite", "down", "during", "except",
+    "for", "from", "in", "inside", "into", "like", "near", "of", "off",
+    "on", "onto", "out", "outside", "over", "past", "since", "through",
+    "throughout", "to", "toward", "towards", "under", "until", "up",
+    "upon", "with", "within", "without",
+}
+
+COORDINATORS = {"and", "but", "or", "nor", "for", "so", "yet"}
+
+SUBORDINATORS = {
+    "because", "although", "though", "while", "whereas", "if", "unless",
+    "since", "as", "than", "whether", "after", "before", "until", "when",
+    "where", "once", "whenever", "wherever",
+}
+
+WH_WORDS = {
+    "who", "whom", "whose", "what", "which", "when", "where", "why", "how",
+}
+
+PERSONAL_PRONOUNS = {
+    "i", "me", "my", "mine", "myself",
+    "you", "your", "yours", "yourself", "yourselves",
+    "he", "him", "his", "himself",
+    "she", "her", "hers", "herself",
+    "it", "its", "itself",
+    "we", "us", "our", "ours", "ourselves",
+    "they", "them", "their", "theirs", "themselves",
+}
+
+PRONOUN_CATEGORIES: dict[str, set[str]] = {
+    "first_singular": {"i", "me", "my", "mine", "myself"},
+    "first_plural": {"we", "us", "our", "ours", "ourselves"},
+    "second": {"you", "your", "yours", "yourself", "yourselves"},
+    "third_singular": {"he", "him", "his", "himself", "she", "her", "hers", "herself", "it", "its", "itself"},
+    "third_plural": {"they", "them", "their", "theirs", "themselves"},
+    "possessive": {"my", "mine", "your", "yours", "his", "her", "hers", "its", "our", "ours", "their", "theirs"},
+    "reflexive": {"myself", "yourself", "yourselves", "himself", "herself", "itself", "ourselves", "themselves"},
+}
+
+AUXILIARIES = {
+    "am", "is", "are", "was", "were", "be", "been", "being",
+    "have", "has", "had", "having",
+    "do", "does", "did", "doing", "done",
+    "'m", "'re", "'s", "'ve", "'ll", "'d",
+}
+
+MODALS = {
+    "can", "could", "may", "might", "must", "shall", "should", "will",
+    "would", "ought", "need", "dare",
+}
+
+NEGATIONS = {
+    "not", "no", "never", "none", "nothing", "nobody", "nowhere",
+    "neither", "nor", "n't", "cannot",
+    "can't", "won't", "don't", "doesn't", "didn't", "isn't", "aren't",
+    "wasn't", "weren't", "haven't", "hasn't", "hadn't", "couldn't",
+    "wouldn't", "shouldn't", "mustn't", "ain't",
+}
+
+CONTRACTIONS = {
+    "i'm", "i've", "i'll", "i'd",
+    "you're", "you've", "you'll", "you'd",
+    "he's", "he'll", "he'd",
+    "she's", "she'll", "she'd",
+    "it's", "it'll",
+    "we're", "we've", "we'll", "we'd",
+    "they're", "they've", "they'll", "they'd",
+    "that's", "there's", "what's", "who's", "where's", "how's",
+    "can't", "won't", "don't", "doesn't", "didn't", "isn't", "aren't",
+    "wasn't", "weren't", "haven't", "hasn't", "hadn't", "couldn't",
+    "wouldn't", "shouldn't", "mustn't", "ain't",
+}
+
+DISCOURSE_MARKERS = {
+    "oh", "ah", "eh", "hey", "hi", "hello", "huh", "hmm", "um", "uh",
+    "well", "so", "like", "just", "really", "actually", "anyway",
+    "maybe", "perhaps", "please", "thanks", "thank", "sorry", "wow",
+    "yeah", "yes", "okay", "ok", "right", "still", "even", "only",
+    "quite", "rather", "pretty", "too", "very",
+}
+
+INTERJECTIONS = {
+    "oh", "ah", "eh", "hey", "hi", "hello", "huh", "hmm", "um", "uh",
+    "wow", "yeah", "yes", "okay", "ok", "sigh", "giggle", "sniff",
+    "heh", "haha", "mmm", "phew", "ugh", "ouch", "oops", "yay",
+    "hooray", "bye", "goodbye",
+}
+
+HEDGES = {
+    "maybe", "perhaps", "probably", "possibly", "guess", "suppose",
+    "think", "kind", "sort", "like", "somewhat", "rather", "fairly",
+    "quite",
+}
+
+INTENSIFIERS = {
+    "very", "really", "so", "too", "pretty", "quite", "absolutely",
+    "totally", "completely", "extremely", "incredibly", "terribly",
+    "awfully", "highly",
+}
+
+POLITENESS = {
+    "please", "thanks", "thank", "sorry", "excuse", "pardon", "welcome",
+    "kindly",
+}
+
+# Объединённый список функциональных слов. Он используется и для
+# относительных частот, и для отбора функциональных слов в LSA.
+FUNCTION_WORDS: set[str] = set().union(
+    ARTICLES,
+    DETERMINERS,
+    PREPOSITIONS,
+    COORDINATORS,
+    SUBORDINATORS,
+    WH_WORDS,
+    PERSONAL_PRONOUNS,
+    AUXILIARIES,
+    MODALS,
+    NEGATIONS,
+    CONTRACTIONS,
+)
+
+FUNCTION_WORD_CATEGORIES: dict[str, set[str]] = {
+    "articles": ARTICLES,
+    "determiners": DETERMINERS,
+    "prepositions": PREPOSITIONS,
+    "coordinators": COORDINATORS,
+    "subordinators": SUBORDINATORS,
+    "wh_words": WH_WORDS,
+    "pronouns": PERSONAL_PRONOUNS,
+    "auxiliaries": AUXILIARIES,
+    "modals": MODALS,
+    "negations": NEGATIONS,
+    "contractions": CONTRACTIONS,
+    "discourse_markers": DISCOURSE_MARKERS,
+}
+
+# Запасной список стоп-слов, если NLTK-ресурс stopwords недоступен.
+STOPWORD_FALLBACK = set().union(
+    ARTICLES,
+    DETERMINERS,
+    PREPOSITIONS,
+    COORDINATORS,
+    SUBORDINATORS,
+    WH_WORDS,
+    PERSONAL_PRONOUNS,
+    AUXILIARIES,
+    MODALS,
+    NEGATIONS,
+    {"also", "just", "then", "there", "here", "now", "very", "too", "only"},
+)
+
+# Слова, которые в NLTK помечаются как IN, но синтаксически являются
+# подчинительными союзами. Используются при переходе к Universal POS.
+SUBORDINATING_IN = SUBORDINATORS
diff --git a/lab1/stylometry/lsa.py b/lab1/stylometry/lsa.py
new file mode 100644
index 0000000..a31002e
--- /dev/null
+++ b/lab1/stylometry/lsa.py
@@ -0,0 +1,317 @@
+"""Векторизация текста и латентно-семантический анализ (LSA).
+
+LSA реализован как последовательность:
+TF-IDF -> усечённое сингулярное разложение (TruncatedSVD) -> L2-нормализация.
+Все параметры, включая число компонент, должны настраиваться внутри
+кросс-валидации, иначе возникает утечка информации из тестовой выборки.
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+import numpy as np
+import pandas as pd
+from sklearn.base import BaseEstimator, TransformerMixin
+from sklearn.decomposition import TruncatedSVD
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.preprocessing import normalize
+
+LSA_VARIANTS = ("word", "word2", "content", "fw", "char", "pos")
+
+_DEFAULT_CONFIGS: dict[str, dict[str, Any]] = {
+    "word": {
+        "column": "text",
+        "analyzer": "word",
+        "ngram_range": (1, 1),
+        "token_pattern": r"(?u)\b\w[\w']*\b",
+        "lowercase": True,
+        "min_df": 3,
+        "max_df": 0.9,
+        "max_features": 5000,
+    },
+    "word2": {
+        "column": "text",
+        "analyzer": "word",
+        "ngram_range": (1, 2),
+        "token_pattern": r"(?u)\b\w[\w']*\b",
+        "lowercase": True,
+        "min_df": 3,
+        "max_df": 0.9,
+        "max_features": 8000,
+    },
+    "content": {
+        "column": "content_text",
+        "analyzer": "word",
+        "ngram_range": (1, 2),
+        "token_pattern": r"\S+",
+        "lowercase": True,
+        "min_df": 3,
+        "max_df": 0.9,
+        "max_features": 5000,
+    },
+    "fw": {
+        "column": "fw_text",
+        "analyzer": "word",
+        "ngram_range": (1, 2),
+        "token_pattern": r"\S+",
+        "lowercase": True,
+        "min_df": 3,
+        "max_df": 0.95,
+        "max_features": 2000,
+    },
+    "char": {
+        "column": "text",
+        "analyzer": "char_wb",
+        "ngram_range": (2, 5),
+        "token_pattern": None,
+        "lowercase": True,
+        "min_df": 3,
+        "max_df": 0.95,
+        "max_features": 5000,
+    },
+    "pos": {
+        "column": "pos_text",
+        "analyzer": "word",
+        "ngram_range": (1, 4),
+        "token_pattern": r"\S+",
+        "lowercase": False,
+        "min_df": 3,
+        "max_df": 0.95,
+        "max_features": 3000,
+    },
+}
+
+
+def _as_texts(X: Any, column: str) -> list[str]:
+    if isinstance(X, pd.DataFrame):
+        if column not in X.columns:
+            raise KeyError(f"Нет колонки {column!r} для векторизации.")
+        return X[column].astype(str).tolist()
+    return [str(value) for value in X]
+
+
+class ColumnTfidfVectorizer(BaseEstimator, TransformerMixin):
+    """TF-IDF по одной колонке таблицы признаков."""
+
+    def __init__(
+        self,
+        column: str = "text",
+        analyzer: str = "word",
+        ngram_range: tuple[int, int] = (1, 1),
+        token_pattern: str | None = r"(?u)\b\w[\w']*\b",
+        lowercase: bool = True,
+        min_df: int | float = 1,
+        max_df: int | float = 1.0,
+        max_features: int | None = None,
+        sublinear_tf: bool = True,
+        stop_words: str | None = None,
+    ) -> None:
+        self.column = column
+        self.analyzer = analyzer
+        self.ngram_range = ngram_range
+        self.token_pattern = token_pattern
+        self.lowercase = lowercase
+        self.min_df = min_df
+        self.max_df = max_df
+        self.max_features = max_features
+        self.sublinear_tf = sublinear_tf
+        self.stop_words = stop_words
+
+    def _make_vectorizer(self) -> TfidfVectorizer:
+        kwargs: dict[str, Any] = {
+            "analyzer": self.analyzer,
+            "ngram_range": self.ngram_range,
+            "lowercase": self.lowercase,
+            "min_df": self.min_df,
+            "max_df": self.max_df,
+            "max_features": self.max_features,
+            "sublinear_tf": self.sublinear_tf,
+            "stop_words": self.stop_words,
+        }
+        if self.analyzer == "word" and self.token_pattern is not None:
+            kwargs["token_pattern"] = self.token_pattern
+        return TfidfVectorizer(**kwargs)
+
+    def fit(self, X: Any, y=None) -> "ColumnTfidfVectorizer":
+        texts = _as_texts(X, self.column)
+        if not texts:
+            raise ValueError("Пустая выборка для TF-IDF.")
+        self.vectorizer_ = self._make_vectorizer()
+        self.vectorizer_.fit(texts)
+        self.n_features_in_ = len(self.vectorizer_.get_feature_names_out())
+        return self
+
+    def transform(self, X: Any) -> Any:
+        if not hasattr(self, "vectorizer_"):
+            raise RuntimeError("Сначала вызовите fit.")
+        return self.vectorizer_.transform(_as_texts(X, self.column))
+
+    def get_feature_names_out(self, input_features=None) -> np.ndarray:
+        if not hasattr(self, "vectorizer_"):
+            raise RuntimeError("Сначала вызовите fit.")
+        return self.vectorizer_.get_feature_names_out()
+
+
+class LSAVectorizer(BaseEstimator, TransformerMixin):
+    """TF-IDF + TruncatedSVD + L2-нормализация."""
+
+    def __init__(
+        self,
+        column: str = "text",
+        analyzer: str = "word",
+        ngram_range: tuple[int, int] = (1, 1),
+        token_pattern: str | None = r"(?u)\b\w[\w']*\b",
+        lowercase: bool = True,
+        min_df: int | float = 3,
+        max_df: int | float = 0.9,
+        max_features: int | None = 5000,
+        n_components: int = 100,
+        sublinear_tf: bool = True,
+        stop_words: str | None = None,
+        normalize_rows: bool = True,
+        random_state: int = 42,
+    ) -> None:
+        self.column = column
+        self.analyzer = analyzer
+        self.ngram_range = ngram_range
+        self.token_pattern = token_pattern
+        self.lowercase = lowercase
+        self.min_df = min_df
+        self.max_df = max_df
+        self.max_features = max_features
+        self.n_components = n_components
+        self.sublinear_tf = sublinear_tf
+        self.stop_words = stop_words
+        self.normalize_rows = normalize_rows
+        self.random_state = random_state
+
+    def fit(self, X: Any, y=None) -> "LSAVectorizer":
+        self.tfidf_ = ColumnTfidfVectorizer(
+            column=self.column,
+            analyzer=self.analyzer,
+            ngram_range=self.ngram_range,
+            token_pattern=self.token_pattern,
+            lowercase=self.lowercase,
+            min_df=self.min_df,
+            max_df=self.max_df,
+            max_features=self.max_features,
+            sublinear_tf=self.sublinear_tf,
+            stop_words=self.stop_words,
+        )
+        matrix = self.tfidf_.fit_transform(X)
+        max_components = min(matrix.shape[0] - 1, matrix.shape[1] - 1)
+        if max_components < 1:
+            raise ValueError(
+                "Слишком маленькая матрица для LSA: нужно хотя бы два документа "
+                "и два признака."
+            )
+        self.n_components_ = int(min(self.n_components, max_components))
+        self.svd_ = TruncatedSVD(
+            n_components=self.n_components_,
+            algorithm="randomized",
+            n_iter=10,
+            random_state=self.random_state,
+        )
+        self.svd_.fit(matrix)
+        self.explained_variance_ratio_ = self.svd_.explained_variance_ratio_
+        self.cumulative_explained_variance_ = np.cumsum(self.explained_variance_ratio_)
+        self.n_features_in_ = matrix.shape[1]
+        return self
+
+    def transform(self, X: Any) -> np.ndarray:
+        if not hasattr(self, "svd_"):
+            raise RuntimeError("Сначала вызовите fit.")
+        matrix = self.tfidf_.transform(X)
+        result = self.svd_.transform(matrix)
+        if self.normalize_rows:
+            result = normalize(result, norm="l2", copy=False)
+        return np.asarray(result)
+
+    def fit_transform(self, X: Any, y=None, **fit_params) -> np.ndarray:
+        return self.fit(X, y).transform(X)
+
+    def get_feature_names_out(self, input_features=None) -> np.ndarray:
+        return np.array([f"lsa_{index}" for index in range(self.n_components_)], dtype=object)
+
+    def top_terms(self, n: int = 10) -> list[list[tuple[str, float]]]:
+        """Возвращает топ-``n`` терминов для каждой компоненты.
+
+        Знак компоненты не интерпретируется: он произволен. Для отчёта важнее
+        набор терминов и их относительный вклад.
+        """
+        if not hasattr(self, "svd_"):
+            raise RuntimeError("Сначала вызовите fit.")
+        terms = self.tfidf_.get_feature_names_out()
+        result: list[list[tuple[str, float]]] = []
+        for component in self.svd_.components_:
+            order = np.argsort(component)[::-1][:n]
+            result.append([(str(terms[index]), float(component[index])) for index in order])
+        return result
+
+    def reconstruction_error(self, X: Any) -> float:
+        """Относительная ошибка восстановления TF-IDF матрицы."""
+        matrix = self.tfidf_.transform(X)
+        reconstructed = self.svd_.inverse_transform(self.svd_.transform(matrix))
+        denominator = np.linalg.norm(matrix.toarray())
+        if denominator == 0:
+            return 0.0
+        return float(np.linalg.norm(matrix.toarray() - reconstructed) / denominator)
+
+
+def make_lsa_vectorizer(
+    variant: str = "word",
+    n_components: int = 100,
+    **overrides: Any,
+) -> LSAVectorizer:
+    """Создаёт LSA-векторизатор одного из заранее определённых вариантов."""
+    if variant not in _DEFAULT_CONFIGS:
+        raise ValueError(
+            f"Неизвестный вариант LSA: {variant}. Доступны: {LSA_VARIANTS}"
+        )
+    config = dict(_DEFAULT_CONFIGS[variant])
+    config.update(overrides)
+    return LSAVectorizer(n_components=n_components, **config)
+
+
+def make_text_vectorizer(kind: str = "word", **overrides: Any) -> ColumnTfidfVectorizer:
+    """Создаёт TF-IDF-векторизатор без понижения размерности."""
+    if kind not in _DEFAULT_CONFIGS:
+        raise ValueError(f"Неизвестный вариант TF-IDF: {kind}")
+    config = dict(_DEFAULT_CONFIGS[kind])
+    config.update(overrides)
+    return ColumnTfidfVectorizer(**config)
+
+
+def make_pos_ngram_vectorizer(**overrides: Any) -> ColumnTfidfVectorizer:
+    """TF-IDF по последовательностям Universal POS."""
+    config = {
+        "column": "pos_text",
+        "analyzer": "word",
+        "ngram_range": (1, 3),
+        "token_pattern": r"\S+",
+        "lowercase": False,
+        "min_df": 3,
+        "max_df": 0.95,
+        "max_features": 500,
+        "sublinear_tf": True,
+    }
+    config.update(overrides)
+    return ColumnTfidfVectorizer(**config)
+
+
+def make_char_ngram_vectorizer(**overrides: Any) -> ColumnTfidfVectorizer:
+    """TF-IDF по символьным n-граммам."""
+    config = {
+        "column": "text",
+        "analyzer": "char_wb",
+        "ngram_range": (2, 5),
+        "token_pattern": None,
+        "lowercase": True,
+        "min_df": 3,
+        "max_df": 0.95,
+        "max_features": 5000,
+        "sublinear_tf": True,
+    }
+    config.update(overrides)
+    return ColumnTfidfVectorizer(**config)
diff --git a/lab1/stylometry/metadata.py b/lab1/stylometry/metadata.py
new file mode 100644
index 0000000..6a4c9bf
--- /dev/null
+++ b/lab1/stylometry/metadata.py
@@ -0,0 +1,99 @@
+"""Анализ ошибок по метаданным персонажей."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Sequence
+
+import pandas as pd
+
+from .evaluation import per_class_report
+
+DEFAULT_METADATA_PATH = Path(__file__).resolve().parents[1] / "data" / "metadata" / "characters.csv"
+
+METADATA_COLUMNS = ("gender", "age_group", "species", "role")
+
+
+def load_character_metadata(path: str | Path | None = None) -> pd.DataFrame:
+    """Читает таблицу пола, возраста, вида и роли персонажей."""
+    path = Path(path) if path is not None else DEFAULT_METADATA_PATH
+    if not path.exists():
+        raise FileNotFoundError(f"Нет файла метаданных: {path}")
+    frame = pd.read_csv(path)
+    required = {"character", *METADATA_COLUMNS}
+    missing = required - set(frame.columns)
+    if missing:
+        raise ValueError(f"В метаданных нет колонок: {sorted(missing)}")
+    frame = frame.copy()
+    frame["character"] = frame["character"].astype(str).str.strip()
+    return frame
+
+
+def subgroup_report(
+    y_true: Sequence,
+    y_pred: Sequence,
+    labels: Sequence,
+    metadata: pd.DataFrame,
+    group_column: str = "gender",
+) -> pd.DataFrame:
+    """Сводка F-меры по подгруппам персонажей."""
+    if group_column not in metadata.columns:
+        raise ValueError(f"Нет колонки {group_column!r} в метаданных.")
+    per_class = per_class_report(y_true, y_pred, labels)
+    per_class = per_class.merge(
+        metadata[["character", group_column]],
+        left_on="class",
+        right_on="character",
+        how="left",
+    )
+    rows = []
+    for value, group in per_class.groupby(group_column, dropna=False):
+        support = int(group["support"].sum())
+        weighted = float((group["f1"] * group["support"]).sum() / max(support, 1))
+        rows.append(
+            {
+                "group_column": group_column,
+                "group": str(value),
+                "n_characters": int(len(group)),
+                "n_samples": support,
+                "weighted_f1": weighted,
+                "mean_class_f1": float(group["f1"].mean()),
+                "min_class_f1": float(group["f1"].min()),
+                "max_class_f1": float(group["f1"].max()),
+            }
+        )
+    return pd.DataFrame(rows).sort_values(["weighted_f1"], ascending=False)
+
+
+def group_confusions(
+    y_true: Sequence,
+    y_pred: Sequence,
+    metadata: pd.DataFrame,
+    group_column: str = "gender",
+) -> pd.DataFrame:
+    """Считает, как часто ошибки происходят внутри и между группами."""
+    frame = pd.DataFrame({"true_character": list(y_true), "pred_character": list(y_pred)})
+    frame = frame.merge(
+        metadata[["character", group_column]].rename(
+            columns={"character": "true_character", group_column: "true_group"}
+        ),
+        on="true_character",
+        how="left",
+    ).merge(
+        metadata[["character", group_column]].rename(
+            columns={"character": "pred_character", group_column: "pred_group"}
+        ),
+        on="pred_character",
+        how="left",
+    )
+    errors = frame[frame["true_character"] != frame["pred_character"]]
+    table = (
+        errors.groupby(["true_group", "pred_group"], dropna=False)
+        .size()
+        .reset_index(name="count")
+        .sort_values("count", ascending=False)
+    )
+    total = max(int(table["count"].sum()), 1)
+    table["share_of_errors"] = table["count"] / total
+    table["same_group"] = table["true_group"] == table["pred_group"]
+    return table
diff --git a/lab1/stylometry/pipelines.py b/lab1/stylometry/pipelines.py
new file mode 100644
index 0000000..47bb604
--- /dev/null
+++ b/lab1/stylometry/pipelines.py
@@ -0,0 +1,365 @@
+"""Сборка признаковых пространств и классификаторов."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+from sklearn.dummy import DummyClassifier
+from sklearn.ensemble import ExtraTreesClassifier, RandomForestClassifier
+from sklearn.linear_model import LogisticRegression
+from sklearn.model_selection import GridSearchCV
+from sklearn.naive_bayes import MultinomialNB
+from sklearn.neighbors import KNeighborsClassifier
+from sklearn.pipeline import FeatureUnion, Pipeline
+from sklearn.preprocessing import MinMaxScaler, Normalizer, StandardScaler
+from sklearn.svm import LinearSVC
+
+from .features import FEATURE_GROUPS, StylometricVectorizer
+from .lsa import (
+    LSA_VARIANTS,
+    make_char_ngram_vectorizer,
+    make_lsa_vectorizer,
+    make_pos_ngram_vectorizer,
+)
+
+MODEL_NAMES = (
+    "dummy",
+    "multinomial_nb",
+    "logreg",
+    "linear_svc",
+    "random_forest",
+    "extra_trees",
+    "knn",
+)
+
+
+@dataclass
+class FeatureConfig:
+    """Описание признакового пространства."""
+
+    groups: tuple[str, ...] = field(default_factory=lambda: tuple(FEATURE_GROUPS))
+    use_char_ngrams: bool = True
+    use_pos_ngrams: bool = True
+    lsa_variants: tuple[str, ...] = ("word", "pos")
+    lsa_components: int = 100
+    stylo_scaler: str = "standard"  # standard | minmax | none
+    stylo_normalize: bool = True
+    include_stylo: bool = True
+    include_lsa: bool = True
+    char_max_features: int = 5000
+    pos_max_features: int = 500
+    char_ngram_range: tuple[int, int] = (2, 5)
+    pos_ngram_range: tuple[int, int] = (1, 3)
+    min_df: int = 3
+    random_state: int = 42
+
+
+def _stylo_branch(config: FeatureConfig) -> Pipeline:
+    if config.stylo_scaler == "standard":
+        scaler: Any = StandardScaler()
+    elif config.stylo_scaler == "minmax":
+        scaler = MinMaxScaler()
+    elif config.stylo_scaler == "none":
+        scaler = "passthrough"
+    else:
+        raise ValueError(f"Неизвестный масштабировщик: {config.stylo_scaler}")
+    steps: list[tuple[str, Any]] = [
+        ("extract", StylometricVectorizer(groups=config.groups)),
+        ("scale", scaler),
+    ]
+    if config.stylo_normalize:
+        # L2-нормировка уравнивает вклад блоков разной размерности.
+        steps.append(("norm", Normalizer()))
+    return Pipeline(steps)
+
+
+def build_feature_union(config: FeatureConfig) -> Any:
+    """Собирает блоки признаков в единое пространство."""
+    branches: list[tuple[str, Any]] = []
+
+    if config.include_stylo:
+        branches.append(("stylo", _stylo_branch(config)))
+
+    if config.use_char_ngrams:
+        branches.append(
+            (
+                "char",
+                Pipeline(
+                    [
+                        (
+                            "tfidf",
+                            make_char_ngram_vectorizer(
+                                ngram_range=config.char_ngram_range,
+                                max_features=config.char_max_features,
+                                min_df=config.min_df,
+                            ),
+                        ),
+                        ("norm", Normalizer()),
+                    ]
+                ),
+            )
+        )
+
+    if config.use_pos_ngrams:
+        branches.append(
+            (
+                "pos",
+                Pipeline(
+                    [
+                        (
+                            "tfidf",
+                            make_pos_ngram_vectorizer(
+                                ngram_range=config.pos_ngram_range,
+                                max_features=config.pos_max_features,
+                                min_df=config.min_df,
+                            ),
+                        ),
+                        ("norm", Normalizer()),
+                    ]
+                ),
+            )
+        )
+
+    if config.include_lsa:
+        for variant in config.lsa_variants:
+            if variant not in LSA_VARIANTS:
+                raise ValueError(f"Неизвестный вариант LSA: {variant}")
+            branches.append(
+                (
+                    f"lsa_{variant}",
+                    Pipeline(
+                        [
+                            (
+                                "lsa",
+                                make_lsa_vectorizer(
+                                    variant=variant,
+                                    n_components=config.lsa_components,
+                                    min_df=config.min_df,
+                                    random_state=config.random_state,
+                                ),
+                            ),
+                            ("norm", Normalizer()),
+                        ]
+                    ),
+                )
+            )
+
+    if not branches:
+        raise ValueError("Не выбран ни один блок признаков.")
+
+    if len(branches) == 1:
+        return branches[0][1]
+    return FeatureUnion(branches)
+
+
+def make_classifier(name: str, **params: Any) -> Any:
+    """Создаёт классификатор по короткому имени."""
+    if name == "dummy":
+        strategy = params.pop("strategy", "most_frequent")
+        return DummyClassifier(strategy=strategy)
+    if name == "multinomial_nb":
+        return MultinomialNB(alpha=params.pop("alpha", 1.0))
+    if name == "logreg":
+        return LogisticRegression(
+            C=params.pop("C", 1.0),
+            solver="lbfgs",
+            max_iter=5000,
+            class_weight="balanced",
+        )
+    if name == "linear_svc":
+        return LinearSVC(
+            C=params.pop("C", 1.0),
+            class_weight="balanced",
+            max_iter=20_000,
+            dual="auto",
+        )
+    if name == "random_forest":
+        return RandomForestClassifier(
+            n_estimators=params.pop("n_estimators", 400),
+            max_depth=params.pop("max_depth", None),
+            min_samples_leaf=params.pop("min_samples_leaf", 1),
+            class_weight="balanced_subsample",
+            n_jobs=-1,
+            random_state=params.pop("random_state", 42),
+        )
+    if name == "extra_trees":
+        return ExtraTreesClassifier(
+            n_estimators=params.pop("n_estimators", 400),
+            max_depth=params.pop("max_depth", None),
+            min_samples_leaf=params.pop("min_samples_leaf", 1),
+            class_weight="balanced",
+            n_jobs=-1,
+            random_state=params.pop("random_state", 42),
+        )
+    if name == "knn":
+        return KNeighborsClassifier(
+            n_neighbors=params.pop("n_neighbors", 5),
+            metric=params.pop("metric", "cosine"),
+            weights=params.pop("weights", "distance"),
+            n_jobs=-1,
+        )
+    raise ValueError(f"Неизвестный классификатор: {name}. Доступны: {MODEL_NAMES}")
+
+
+def build_pipeline(
+    feature_config: FeatureConfig,
+    model_name: str,
+    **model_params: Any,
+) -> Pipeline:
+    """Собирает полный конвейер: признаки -> классификатор.
+
+    Для MultinomialNB исключаются LSA-компоненты (они могут быть
+    отрицательными) и используется MinMax-масштабирование, чтобы матрица
+    оставалась неотрицательной.
+    """
+    if model_name == "dummy":
+        return Pipeline([("model", make_classifier(model_name, **model_params))])
+
+    if model_name == "multinomial_nb":
+        config = FeatureConfig(
+            groups=feature_config.groups,
+            use_char_ngrams=feature_config.use_char_ngrams,
+            use_pos_ngrams=feature_config.use_pos_ngrams,
+            lsa_variants=(),
+            lsa_components=feature_config.lsa_components,
+            stylo_scaler="minmax",
+            stylo_normalize=feature_config.stylo_normalize,
+            include_stylo=feature_config.include_stylo,
+            include_lsa=False,
+            char_max_features=feature_config.char_max_features,
+            pos_max_features=feature_config.pos_max_features,
+            char_ngram_range=feature_config.char_ngram_range,
+            pos_ngram_range=feature_config.pos_ngram_range,
+            min_df=feature_config.min_df,
+            random_state=feature_config.random_state,
+        )
+        features = build_feature_union(config)
+        return Pipeline([("features", features), ("model", make_classifier(model_name, **model_params))])
+
+    features = build_feature_union(feature_config)
+    return Pipeline([("features", features), ("model", make_classifier(model_name, **model_params))])
+
+
+def preset_config(name: str = "stylometry_pos_lsa") -> FeatureConfig:
+    """Готовые наборы признаков для типовых экспериментов."""
+    all_groups = tuple(FEATURE_GROUPS)
+    if name == "stylometry":
+        return FeatureConfig(
+            groups=all_groups,
+            use_char_ngrams=False,
+            use_pos_ngrams=False,
+            lsa_variants=(),
+            stylo_normalize=False,
+            include_lsa=False,
+        )
+    if name == "stylometry_pos":
+        return FeatureConfig(
+            groups=all_groups,
+            use_char_ngrams=True,
+            use_pos_ngrams=True,
+            lsa_variants=(),
+            include_lsa=False,
+        )
+    if name in {"stylometry_lsa", "stylometry_pos_lsa"}:
+        return FeatureConfig(
+            groups=all_groups,
+            use_char_ngrams=True,
+            use_pos_ngrams=True,
+            lsa_variants=("word", "pos"),
+            include_lsa=True,
+        )
+    if name == "all":
+        return FeatureConfig(
+            groups=all_groups,
+            use_char_ngrams=True,
+            use_pos_ngrams=True,
+            lsa_variants=("word", "word2", "content", "fw", "char", "pos"),
+            include_lsa=True,
+        )
+    if name == "char":
+        return FeatureConfig(
+            groups=(),
+            use_char_ngrams=True,
+            use_pos_ngrams=False,
+            lsa_variants=(),
+            include_stylo=False,
+            include_lsa=False,
+        )
+    if name == "pos":
+        return FeatureConfig(
+            groups=(),
+            use_char_ngrams=False,
+            use_pos_ngrams=True,
+            lsa_variants=(),
+            include_stylo=False,
+            include_lsa=False,
+        )
+    if name == "fw":
+        return FeatureConfig(
+            groups=("function_words", "pronouns"),
+            use_char_ngrams=False,
+            use_pos_ngrams=False,
+            lsa_variants=(),
+            stylo_normalize=False,
+            include_lsa=False,
+        )
+    if name in {
+        "lsa_word", "lsa_word2", "lsa_content", "lsa_fw", "lsa_char", "lsa_pos"
+    }:
+        variant = name.removeprefix("lsa_")
+        return FeatureConfig(
+            groups=(),
+            use_char_ngrams=False,
+            use_pos_ngrams=False,
+            lsa_variants=(variant,),
+            include_stylo=False,
+            include_lsa=True,
+        )
+    if name == "length":
+        return FeatureConfig(
+            groups=("length",),
+            use_char_ngrams=False,
+            use_pos_ngrams=False,
+            lsa_variants=(),
+            stylo_normalize=False,
+            include_lsa=False,
+        )
+    raise ValueError(f"Неизвестный набор признаков: {name}")
+
+
+PARAM_GRIDS: dict[str, dict[str, list[Any]]] = {
+    "logreg": {"model__C": [0.1, 1.0, 10.0]},
+    "linear_svc": {"model__C": [0.1, 1.0, 10.0]},
+    "random_forest": {
+        "model__max_depth": [None, 30],
+        "model__min_samples_leaf": [1, 2],
+    },
+    "extra_trees": {
+        "model__max_depth": [None, 30],
+        "model__min_samples_leaf": [1, 2],
+    },
+    "multinomial_nb": {"model__alpha": [0.1, 1.0, 3.0]},
+    "knn": {"model__n_neighbors": [3, 5, 9]},
+}
+
+
+def make_grid_search(
+    pipeline: Pipeline,
+    model_name: str,
+    cv: int = 3,
+    scoring: str = "f1_macro",
+    n_jobs: int = -1,
+) -> GridSearchCV:
+    """GridSearchCV с сеткой, специфичной для модели."""
+    if model_name not in PARAM_GRIDS:
+        raise ValueError(f"Для модели {model_name} нет сетки гиперпараметров.")
+    return GridSearchCV(
+        pipeline,
+        PARAM_GRIDS[model_name],
+        cv=cv,
+        scoring=scoring,
+        n_jobs=n_jobs,
+        refit=True,
+        return_train_score=False,
+    )
diff --git a/lab1/stylometry/plotting.py b/lab1/stylometry/plotting.py
new file mode 100644
index 0000000..3e6108a
--- /dev/null
+++ b/lab1/stylometry/plotting.py
@@ -0,0 +1,167 @@
+"""Построение графиков для отчёта и презентации."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Sequence
+
+import matplotlib
+
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt  # noqa: E402
+import numpy as np  # noqa: E402
+import pandas as pd  # noqa: E402
+from sklearn.decomposition import PCA  # noqa: E402
+from sklearn.manifold import TSNE  # noqa: E402
+
+
+def _save(fig: plt.Figure, path: str | Path) -> Path:
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    fig.tight_layout()
+    fig.savefig(path, dpi=200, bbox_inches="tight")
+    plt.close(fig)
+    return path
+
+
+def plot_class_distribution(
+    frame: pd.DataFrame,
+    character_column: str = "character",
+    path: str | Path = "results/figures/class_distribution.png",
+) -> Path:
+    counts = frame[character_column].value_counts().sort_values()
+    fig, ax = plt.subplots(figsize=(8, max(6, len(counts) * 0.25)))
+    ax.barh(counts.index, counts.values)
+    ax.set_xlabel("Число реплик")
+    ax.set_ylabel("Персонаж")
+    ax.set_title("Распределение реплик по персонажам")
+    return _save(fig, path)
+
+
+def plot_scree(
+    explained_variance_ratio: Sequence[float],
+    path: str | Path = "results/figures/lsa_scree.png",
+    title: str = "Доля объяснённой дисперсии компонент LSA",
+) -> Path:
+    values = np.asarray(explained_variance_ratio, dtype=float)
+    cumulative = np.cumsum(values)
+    fig, ax = plt.subplots(figsize=(7, 4.5))
+    ax.plot(np.arange(1, len(values) + 1), cumulative, marker="o", markersize=3)
+    ax.set_xlabel("Число компонент")
+    ax.set_ylabel("Накопленная доля дисперсии")
+    ax.set_title(title)
+    ax.grid(alpha=0.3)
+    return _save(fig, path)
+
+
+def plot_confusion_matrix(
+    matrix: pd.DataFrame,
+    path: str | Path = "results/figures/confusion_matrix.png",
+    title: str = "Матрица ошибок",
+    annotate: bool | None = None,
+) -> Path:
+    values = matrix.to_numpy(dtype=float)
+    n_classes = values.shape[0]
+    if annotate is None:
+        annotate = n_classes <= 20
+    fig, ax = plt.subplots(figsize=(max(8, n_classes * 0.35), max(7, n_classes * 0.32)))
+    image = ax.imshow(values, cmap="Blues", aspect="auto")
+    ax.set_xticks(np.arange(n_classes))
+    ax.set_yticks(np.arange(n_classes))
+    ax.set_xticklabels(matrix.columns, rotation=90, fontsize=7)
+    ax.set_yticklabels(matrix.index, fontsize=7)
+    ax.set_xlabel("Предсказанный класс")
+    ax.set_ylabel("Истинный класс")
+    ax.set_title(title)
+    if annotate:
+        for row in range(n_classes):
+            for column in range(n_classes):
+                value = values[row, column]
+                if value:
+                    ax.text(
+                        column, row, f"{value:.2f}" if value < 1 else f"{int(value)}",
+                        ha="center", va="center", fontsize=6,
+                        color="white" if value > values.max() * 0.5 else "black",
+                    )
+    fig.colorbar(image, ax=ax, fraction=0.046, pad=0.04)
+    return _save(fig, path)
+
+
+def plot_pca(
+    matrix: np.ndarray,
+    labels: Sequence[str],
+    path: str | Path = "results/figures/pca.png",
+    title: str = "PCA признакового пространства",
+) -> Path:
+    values = np.asarray(matrix, dtype=float)
+    if values.shape[0] < 2:
+        raise ValueError("Для PCA нужно хотя бы два объекта.")
+    coordinates = PCA(n_components=2, random_state=42).fit_transform(values)
+    fig, ax = plt.subplots(figsize=(9, 7))
+    unique_labels = sorted(set(labels))
+    cmap = plt.get_cmap("tab20")
+    color_map = {label: cmap(index % 20) for index, label in enumerate(unique_labels)}
+    for label in unique_labels:
+        mask = np.asarray(labels) == label
+        ax.scatter(
+            coordinates[mask, 0], coordinates[mask, 1],
+            s=12, alpha=0.7, label=label, color=color_map[label],
+        )
+    ax.set_xlabel("Первая главная компонента")
+    ax.set_ylabel("Вторая главная компонента")
+    ax.set_title(title)
+    ax.legend(fontsize=6, ncol=2, loc="best", frameon=False)
+    return _save(fig, path)
+
+
+def plot_tsne(
+    matrix: np.ndarray,
+    labels: Sequence[str],
+    path: str | Path = "results/figures/tsne.png",
+    title: str = "t-SNE признакового пространства",
+    perplexity: float = 30.0,
+) -> Path:
+    values = np.asarray(matrix, dtype=float)
+    if values.shape[0] < 10:
+        raise ValueError("Для t-SNE нужно хотя бы десять объектов.")
+    perplexity = min(perplexity, max(5.0, values.shape[0] / 4.0))
+    coordinates = TSNE(
+        n_components=2,
+        perplexity=perplexity,
+        init="pca",
+        learning_rate="auto",
+        random_state=42,
+    ).fit_transform(values)
+    fig, ax = plt.subplots(figsize=(9, 7))
+    unique_labels = sorted(set(labels))
+    cmap = plt.get_cmap("tab20")
+    color_map = {label: cmap(index % 20) for index, label in enumerate(unique_labels)}
+    for label in unique_labels:
+        mask = np.asarray(labels) == label
+        ax.scatter(
+            coordinates[mask, 0], coordinates[mask, 1],
+            s=10, alpha=0.7, label=label, color=color_map[label],
+        )
+    ax.set_xlabel("t-SNE 1")
+    ax.set_ylabel("t-SNE 2")
+    ax.set_title(title)
+    ax.legend(fontsize=6, ncol=2, loc="best", frameon=False)
+    return _save(fig, path)
+
+
+def plot_feature_importance(
+    names: Sequence[str],
+    values: Sequence[float],
+    path: str | Path = "results/figures/feature_importance.png",
+    title: str = "Важность признаков",
+    top_n: int = 25,
+) -> Path:
+    frame = pd.DataFrame({"feature": list(names), "value": np.asarray(values, dtype=float)})
+    frame["abs_value"] = frame["value"].abs()
+    frame = frame.sort_values("abs_value", ascending=False).head(top_n).iloc[::-1]
+    fig, ax = plt.subplots(figsize=(9, max(5, len(frame) * 0.28)))
+    ax.barh(frame["feature"], frame["value"])
+    ax.set_xlabel("Вклад признака")
+    ax.set_ylabel("Признак")
+    ax.set_title(title)
+    return _save(fig, path)
diff --git a/lab1/stylometry/stats.py b/lab1/stylometry/stats.py
new file mode 100644
index 0000000..2c3c824
--- /dev/null
+++ b/lab1/stylometry/stats.py
@@ -0,0 +1,143 @@
+"""Статистические критерии для сравнения моделей."""
+
+from __future__ import annotations
+
+from typing import Sequence
+
+import numpy as np
+from scipy.stats import binomtest, friedmanchisquare, wilcoxon
+from sklearn.metrics import accuracy_score, f1_score
+
+
+def mcnemar_exact(
+    y_true: Sequence,
+    predictions_a: Sequence,
+    predictions_b: Sequence,
+) -> dict[str, float]:
+    """Точный критерий Мак-Немара для двух парных предсказаний."""
+    y_true = np.asarray(y_true)
+    correct_a = np.asarray(predictions_a) == y_true
+    correct_b = np.asarray(predictions_b) == y_true
+    b = int(np.sum(correct_a & ~correct_b))
+    c = int(np.sum(~correct_a & correct_b))
+    n = b + c
+    p_value = 1.0 if n == 0 else float(binomtest(min(b, c), n, 0.5).pvalue)
+    return {
+        "b": b,
+        "c": c,
+        "n_discordant": n,
+        "statistic": float(min(b, c)),
+        "p_value": p_value,
+    }
+
+
+def wilcoxon_signed_rank(
+    scores_a: Sequence[float],
+    scores_b: Sequence[float],
+) -> dict[str, float]:
+    """Парный критерий Уилкоксона для результатов по фолдам."""
+    scores_a = np.asarray(scores_a, dtype=float)
+    scores_b = np.asarray(scores_b, dtype=float)
+    differences = scores_a - scores_b
+    if np.allclose(differences, 0):
+        return {"statistic": 0.0, "p_value": 1.0, "n": int(len(scores_a))}
+    statistic, p_value = wilcoxon(scores_a, scores_b, alternative="two-sided")
+    return {
+        "statistic": float(statistic),
+        "p_value": float(p_value),
+        "n": int(len(scores_a)),
+    }
+
+
+def friedman_test(*score_arrays: Sequence[float]) -> dict[str, float]:
+    """Критерий Фридмана для сравнения нескольких моделей."""
+    arrays = [np.asarray(values, dtype=float) for values in score_arrays]
+    if len(arrays) < 3:
+        raise ValueError("Для критерия Фридмана нужно хотя бы три модели.")
+    if len({len(values) for values in arrays}) != 1:
+        raise ValueError("Все массивы должны быть одинаковой длины.")
+    statistic, p_value = friedmanchisquare(*arrays)
+    return {"statistic": float(statistic), "p_value": float(p_value)}
+
+
+def holm_bonferroni(p_values: Sequence[float]) -> list[float]:
+    """Поправка Холма—Бонферрони для множественных сравнений."""
+    values = np.asarray(p_values, dtype=float)
+    n = len(values)
+    order = np.argsort(values)
+    adjusted = np.empty(n, dtype=float)
+    previous = 0.0
+    for rank, index in enumerate(order):
+        value = min(1.0, values[index] * (n - rank))
+        previous = max(previous, value)
+        adjusted[index] = previous
+    return adjusted.tolist()
+
+
+def bootstrap_difference(
+    y_true: Sequence,
+    predictions_a: Sequence,
+    predictions_b: Sequence,
+    metric: str = "macro_f1",
+    labels: Sequence | None = None,
+    n_boot: int = 1000,
+    seed: int = 42,
+) -> dict[str, float]:
+    """Доверительный интервал разницы метрик двух моделей."""
+    y_true = np.asarray(y_true)
+    predictions_a = np.asarray(predictions_a)
+    predictions_b = np.asarray(predictions_b)
+    labels_array = np.asarray(labels) if labels is not None else np.unique(y_true)
+
+    def _metric(true_values: np.ndarray, predicted_values: np.ndarray) -> float:
+        if metric == "macro_f1":
+            return float(
+                f1_score(
+                    true_values,
+                    predicted_values,
+                    labels=labels_array,
+                    average="macro",
+                    zero_division=0,
+                )
+            )
+        if metric == "accuracy":
+            return float(accuracy_score(true_values, predicted_values))
+        raise ValueError(f"Неизвестная метрика: {metric}")
+
+    rng = np.random.default_rng(seed)
+    differences = []
+    n_samples = len(y_true)
+    for _ in range(n_boot):
+        indices = rng.integers(0, n_samples, n_samples)
+        differences.append(
+            _metric(y_true[indices], predictions_a[indices])
+            - _metric(y_true[indices], predictions_b[indices])
+        )
+    values = np.asarray(differences)
+    return {
+        "mean_difference": float(values.mean()),
+        "lower": float(np.percentile(values, 2.5)),
+        "upper": float(np.percentile(values, 97.5)),
+        "n_boot": int(n_boot),
+    }
+
+
+def cochran_q(correctness_matrix: np.ndarray) -> dict[str, float]:
+    """Критерий Кокрана для сравнения нескольких бинарных классификаторов.
+
+    ``correctness_matrix`` имеет форму (объекты, модели), значения 0/1.
+    """
+    matrix = np.asarray(correctness_matrix, dtype=float)
+    if matrix.ndim != 2:
+        raise ValueError("Ожидается матрица (объекты, модели).")
+    k = matrix.shape[1]
+    column_sums = matrix.sum(axis=0)
+    row_sums = matrix.sum(axis=1)
+    total = matrix.sum()
+    numerator = (k - 1) * (k * np.sum(column_sums**2) - total**2)
+    denominator = k * total - np.sum(row_sums**2)
+    statistic = 0.0 if denominator == 0 else float(numerator / denominator)
+    from scipy.stats import chi2
+
+    p_value = float(chi2.sf(statistic, df=k - 1))
+    return {"statistic": statistic, "p_value": p_value, "df": k - 1}
diff --git a/lab1/stylometry/tokenization.py b/lab1/stylometry/tokenization.py
new file mode 100644
index 0000000..fa5bfc4
--- /dev/null
+++ b/lab1/stylometry/tokenization.py
@@ -0,0 +1,363 @@
+"""Токенизация, разбиение на предложения и частеречная разметка.
+
+По умолчанию используется NLTK: ``word_tokenize``, ``sent_tokenize`` и
+``pos_tag``. Если ресурсы NLTK недоступны, включается упрощённый резервный
+режим на регулярных выражениях, а POS-признаки отключаются. Это позволяет
+запускать аудит и часть признаков даже без скачанных моделей.
+"""
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Sequence
+
+import pandas as pd
+
+from .lexicons import (
+    AUXILIARIES,
+    FUNCTION_WORDS,
+    MODALS,
+    NEGATIONS,
+    PERSONAL_PRONOUNS,
+    SUBORDINATORS,
+    STOPWORD_FALLBACK,
+)
+
+DEFAULT_NLTK_DIR = Path(__file__).resolve().parents[1] / "nltk_data"
+
+_WORD_RE = re.compile(r"[A-Za-z]+(?:['’][A-Za-z]+)?|\d+|[^\w\s]")
+_SENTENCE_RE = re.compile(r"(?<=[.!?])\s+")
+
+_PUNCT_TAGS = {".", ",", ":", "``", "''", "-LRB-", "-RRB-", "HYPH", "NFP"}
+_PTB_TAGS = [
+    "CC", "CD", "DT", "EX", "FW", "IN", "JJ", "JJR", "JJS", "LS", "MD",
+    "NN", "NNS", "NNP", "NNPS", "PDT", "POS", "PRP", "PRP$", "RB",
+    "RBR", "RBS", "RP", "SYM", "TO", "UH", "VB", "VBD", "VBG", "VBN",
+    "VBP", "VBZ", "WDT", "WP", "WP$", "WRB",
+]
+UPOS_TAGS = [
+    "ADJ", "ADP", "ADV", "AUX", "CONJ", "DET", "INTJ", "NOUN", "NUM",
+    "PRON", "PRT", "PUNCT", "SCONJ", "SYM", "VERB", "X",
+]
+
+_AUX_WORDS = AUXILIARIES | MODALS | {"be", "been", "being", "am", "is", "are", "was", "were"}
+_NEGATION_WORDS = NEGATIONS | {"not", "n't"}
+
+
+def _check_nltk_resource(resource: str) -> bool:
+    try:
+        import nltk
+
+        nltk.data.find(resource)
+        return True
+    except Exception:
+        return False
+
+
+def ensure_nltk_data(
+    nltk_data_dir: str | Path | None = None,
+    download: bool = False,
+) -> bool:
+    """Добавляет локальный каталог NLTK в путь поиска и проверяет ресурсы.
+
+    Возвращает ``True``, если доступны и токенизатор, и POS-теггер.
+    """
+    import nltk
+
+    directory = Path(nltk_data_dir) if nltk_data_dir is not None else DEFAULT_NLTK_DIR
+    if directory.exists():
+        path = str(directory.resolve())
+        if path not in nltk.data.path:
+            nltk.data.path.insert(0, path)
+
+    required = {
+        "tokenizer": "tokenizers/punkt_tab",
+        "tagger": "taggers/averaged_perceptron_tagger_eng",
+    }
+    if download:
+        for package in (
+            "punkt",
+            "punkt_tab",
+            "averaged_perceptron_tagger",
+            "averaged_perceptron_tagger_eng",
+            "stopwords",
+        ):
+            nltk.download(package, download_dir=str(directory), quiet=True)
+
+    return all(_check_nltk_resource(resource) for resource in required.values())
+
+
+def load_stopwords(nltk_data_dir: str | Path | None = None) -> set[str]:
+    """Загружает английские стоп-слова NLTK или использует запасной список."""
+    try:
+        import nltk
+
+        directory = Path(nltk_data_dir) if nltk_data_dir is not None else DEFAULT_NLTK_DIR
+        if directory.exists():
+            path = str(directory.resolve())
+            if path not in nltk.data.path:
+                nltk.data.path.insert(0, path)
+        return set(nltk.corpus.stopwords.words("english"))
+    except Exception:
+        return set(STOPWORD_FALLBACK)
+
+
+def penn_to_upos(token: str, tag: str) -> str:
+    """Переводит тег Penn Treebank в Universal POS.
+
+    Часть решений зависит не только от тега, но и от самого слова: например,
+    ``do`` может быть смысловым глаголом или вспомогательным, а ``as`` —
+    предлогом или подчинительным союзом.
+    """
+    lower = token.lower()
+    if tag in _PUNCT_TAGS or token in {".", ",", "!", "?", ";", ":", "...", "…"}:
+        return "PUNCT"
+    if tag == "SYM":
+        return "SYM"
+    if tag == "CC":
+        return "CONJ"
+    if tag == "IN":
+        return "SCONJ" if lower in SUBORDINATORS else "ADP"
+    if tag == "TO":
+        return "PRT"
+    if tag == "RP" or tag == "POS":
+        return "PRT"
+    if tag == "MD":
+        return "AUX"
+    if tag == "EX":
+        return "PRON"
+    if tag in {"WDT", "WP", "WP$"}:
+        return "PRON"
+    if tag == "WRB":
+        return "ADV"
+    if tag == "UH":
+        return "INTJ"
+    if tag == "CD":
+        return "NUM"
+    if tag in {"DT", "PDT"}:
+        return "DET"
+    if tag in {"PRP", "PRP$"}:
+        return "PRON"
+    if tag.startswith("JJ"):
+        return "ADJ"
+    if tag.startswith("RB"):
+        return "ADV"
+    if tag.startswith("NN"):
+        return "NOUN"
+    if tag.startswith("VB"):
+        if lower in _AUX_WORDS:
+            return "AUX"
+        return "VERB"
+    if tag == "FW":
+        return "X"
+    if tag == "LS":
+        return "X"
+
+    # Уточнения по слову для распространённых разговорных форм.
+    if lower in _NEGATION_WORDS:
+        return "PRT"
+    if lower in _AUX_WORDS:
+        return "AUX"
+    if lower in PERSONAL_PRONOUNS:
+        return "PRON"
+    return "X"
+
+
+def collapse_punctuation(tags: Sequence[str]) -> list[str]:
+    """Схлопывает подряд идущие знаки препинания в один тег PUNCT."""
+    result: list[str] = []
+    for tag in tags:
+        if tag == "PUNCT" and result and result[-1] == "PUNCT":
+            continue
+        result.append(tag)
+    return result
+
+
+def is_function_word(token: str, upos: str | None = None) -> bool:
+    """Определяет, относится ли слово к закрытому классу."""
+    lower = token.lower()
+    if lower in FUNCTION_WORDS:
+        return True
+    if upos in {"ADP", "AUX", "CONJ", "DET", "PRON", "PRT", "SCONJ"}:
+        return True
+    return False
+
+
+def is_content_word(token: str, upos: str | None = None, stopwords: set[str] | None = None) -> bool:
+    """Определяет, является ли слово содержательным для content-LSA."""
+    lower = token.lower()
+    if not re.search(r"[A-Za-z]", token):
+        return False
+    if is_function_word(token, upos):
+        return False
+    if stopwords and lower in stopwords:
+        return False
+    return True
+
+
+@dataclass
+class AnalyzedLine:
+    """Результат токенизации и разметки одной реплики."""
+
+    tokens: list[str] = field(default_factory=list)
+    ptb: list[str] = field(default_factory=list)
+    upos: list[str] = field(default_factory=list)
+    sentences: list[list[str]] = field(default_factory=list)
+    pos_text: str = ""
+    fw_text: str = ""
+    content_text: str = ""
+
+
+class Tokenizer:
+    """Токенизатор с NLTK и резервным режимом."""
+
+    def __init__(
+        self,
+        use_nltk: bool = True,
+        nltk_data_dir: str | Path | None = None,
+        strict: bool = False,
+    ) -> None:
+        self.use_nltk = bool(use_nltk)
+        self.nltk_data_dir = Path(nltk_data_dir) if nltk_data_dir is not None else DEFAULT_NLTK_DIR
+        self.strict = strict
+        self.stopwords = load_stopwords(self.nltk_data_dir)
+
+        if self.use_nltk:
+            available = ensure_nltk_data(self.nltk_data_dir)
+            if not available:
+                message = (
+                    "Ресурсы NLTK не найдены. Скачайте punkt_tab и "
+                    "averaged_perceptron_tagger_eng или запустите с --no-nltk."
+                )
+                if self.strict:
+                    raise RuntimeError(message)
+                print(f"[tokenization] {message} Включаю резервный режим.")
+                self.use_nltk = False
+
+        self._pos_available = self.use_nltk and _check_nltk_resource(
+            "taggers/averaged_perceptron_tagger_eng"
+        )
+        self._word_tokenize = None
+        self._sent_tokenize = None
+        self._pos_tag = None
+        if self.use_nltk:
+            from nltk import pos_tag
+            from nltk.tokenize import sent_tokenize, word_tokenize
+
+            self._word_tokenize = word_tokenize
+            self._sent_tokenize = sent_tokenize
+            self._pos_tag = pos_tag
+
+    @property
+    def pos_available(self) -> bool:
+        return self._pos_available
+
+    def tokenize(self, text: str) -> list[str]:
+        if self._word_tokenize is not None:
+            try:
+                return [token for token in self._word_tokenize(text) if token.strip()]
+            except Exception:
+                pass
+        return _WORD_RE.findall(text)
+
+    def sentence_tokenize(self, text: str) -> list[str]:
+        if self._sent_tokenize is not None:
+            try:
+                sentences = [sentence.strip() for sentence in self._sent_tokenize(text)]
+                return [sentence for sentence in sentences if sentence]
+            except Exception:
+                pass
+        return [part.strip() for part in _SENTENCE_RE.split(text) if part.strip()]
+
+    def pos_tag(self, tokens: Sequence[str]) -> list[tuple[str, str]]:
+        if self._pos_tag is not None:
+            try:
+                return list(self._pos_tag(list(tokens)))
+            except Exception:
+                pass
+        return [(token, "X") for token in tokens]
+
+    def analyze(self, text: str) -> AnalyzedLine:
+        tokens = self.tokenize(text)
+        tagged = self.pos_tag(tokens)
+        ptb = [tag for _, tag in tagged]
+        upos = [penn_to_upos(token, tag) for token, tag in tagged]
+
+        sentences: list[list[str]] = []
+        for sentence in self.sentence_tokenize(text):
+            sentence_tokens = self.tokenize(sentence)
+            if sentence_tokens:
+                sentences.append(sentence_tokens)
+        if not sentences and tokens:
+            sentences = [tokens]
+
+        pos_sequence = collapse_punctuation(upos)
+        pos_text = " ".join(pos_sequence) if pos_sequence else "__EMPTY__"
+
+        fw_tokens = [
+            token
+            for token, tag in zip(tokens, upos)
+            if is_function_word(token, tag)
+        ]
+        fw_text = " ".join(fw_tokens) if fw_tokens else "__EMPTY__"
+
+        content_tokens = [
+            token
+            for token, tag in zip(tokens, upos)
+            if is_content_word(token, tag, self.stopwords)
+        ]
+        content_text = " ".join(content_tokens) if content_tokens else "__EMPTY__"
+
+        return AnalyzedLine(
+            tokens=tokens,
+            ptb=ptb,
+            upos=upos,
+            sentences=sentences,
+            pos_text=pos_text,
+            fw_text=fw_text,
+            content_text=content_text,
+        )
+
+    def analyze_frame(
+        self,
+        frame: pd.DataFrame,
+        text_column: str = "text",
+        verbose: bool = False,
+    ) -> pd.DataFrame:
+        """Добавляет к таблице колонки с токенами, тегами и текстами для LSA."""
+        if text_column not in frame.columns:
+            raise KeyError(f"Нет колонки {text_column!r}")
+
+        records: list[dict[str, object]] = []
+        total = len(frame)
+        for position, (_, row) in enumerate(frame.iterrows(), start=1):
+            analyzed = self.analyze(str(row[text_column]))
+            records.append(
+                {
+                    "tokens": analyzed.tokens,
+                    "ptb": analyzed.ptb,
+                    "upos": analyzed.upos,
+                    "sentences": analyzed.sentences,
+                    "pos_text": analyzed.pos_text,
+                    "fw_text": analyzed.fw_text,
+                    "content_text": analyzed.content_text,
+                }
+            )
+            if verbose and position % 500 == 0:
+                print(f"[tokenization] обработано {position} из {total}")
+
+        additions = pd.DataFrame(records, index=frame.index)
+        result = pd.concat([frame.copy(), additions], axis=1)
+        return result
+
+
+def upos_tag_list() -> list[str]:
+    """Список универсальных частеречных тегов, используемых в признаках."""
+    return list(UPOS_TAGS)
+
+
+def ptb_tag_list() -> list[str]:
+    """Список тегов Penn Treebank, используемых в признаках."""
+    return list(_PTB_TAGS)
diff --git a/lab1/stylometry/verification.py b/lab1/stylometry/verification.py
new file mode 100644
index 0000000..a679a5b
--- /dev/null
+++ b/lab1/stylometry/verification.py
@@ -0,0 +1,129 @@
+"""Верификация «тот же персонаж / другой персонаж» по центроидам классов."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, Sequence
+
+import numpy as np
+import pandas as pd
+from scipy.sparse import issparse
+from sklearn.metrics import roc_auc_score, roc_curve
+from sklearn.metrics.pairwise import cosine_similarity
+from sklearn.model_selection import StratifiedKFold
+from sklearn.base import clone
+
+from .pipelines import FeatureConfig, build_feature_union
+
+
+def _to_dense(matrix: Any) -> np.ndarray:
+    return matrix.toarray() if issparse(matrix) else np.asarray(matrix)
+
+
+@dataclass
+class VerificationResult:
+    fold_metrics: pd.DataFrame
+    scores: pd.DataFrame
+    summary: dict[str, float]
+
+
+def equal_error_rate(y_true: Sequence[int], scores: Sequence[float]) -> float:
+    """Равная ошибка: точка, где FPR и FNR совпадают."""
+    fpr, tpr, _ = roc_curve(y_true, scores)
+    fnr = 1.0 - tpr
+    index = int(np.argmin(np.abs(fpr - fnr)))
+    return float((fpr[index] + fnr[index]) / 2.0)
+
+
+def tar_at_far(
+    y_true: Sequence[int], scores: Sequence[float], far: float = 0.01
+) -> float:
+    """Доля принятых «своих» при заданной доле ложных принятий."""
+    fpr, tpr, _ = roc_curve(y_true, scores)
+    return float(np.interp(far, fpr, tpr))
+
+
+def _centroids(matrix: np.ndarray, labels: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
+    classes = np.unique(labels)
+    centroids = np.vstack([matrix[labels == label].mean(axis=0) for label in classes])
+    return classes, centroids
+
+
+def run_verification(
+    X: pd.DataFrame,
+    y: Sequence,
+    feature_config: FeatureConfig,
+    n_splits: int = 5,
+    seed: int = 42,
+    verbose: bool = True,
+) -> VerificationResult:
+    """Строит оценки «свой/чужой» для каждой реплики и каждого персонажа.
+
+    На каждом фолде центроиды персонажей считаются только по обучающим
+    репликам. Для тестовой реплики косинусная близость к центроиду своего
+    персонажа образует положительный класс, к остальным — отрицательный.
+    """
+    y = np.asarray(y)
+    splitter = StratifiedKFold(n_splits=n_splits, shuffle=True, random_state=seed)
+    fold_rows: list[dict[str, Any]] = []
+    score_rows: list[dict[str, Any]] = []
+
+    for fold, (train_idx, test_idx) in enumerate(splitter.split(X, y), start=1):
+        features = build_feature_union(feature_config)
+        features = clone(features)
+        train_matrix = _to_dense(features.fit_transform(X.iloc[train_idx]))
+        test_matrix = _to_dense(features.transform(X.iloc[test_idx]))
+        train_labels = y[train_idx]
+        classes, centroids = _centroids(train_matrix, train_labels)
+        similarities = cosine_similarity(test_matrix, centroids)
+
+        binary_true: list[int] = []
+        binary_scores: list[float] = []
+        top1_correct = 0
+        for row, sample_index in enumerate(test_idx):
+            true_label = y[sample_index]
+            for class_index, class_label in enumerate(classes):
+                score = float(similarities[row, class_index])
+                is_same = int(class_label == true_label)
+                binary_true.append(is_same)
+                binary_scores.append(score)
+                score_rows.append(
+                    {
+                        "fold": fold,
+                        "line_id": int(X.iloc[sample_index]["line_id"]) if "line_id" in X.columns else int(sample_index),
+                        "true_character": true_label,
+                        "candidate_character": class_label,
+                        "score": score,
+                        "is_same": is_same,
+                    }
+                )
+            predicted_label = classes[int(np.argmax(similarities[row]))]
+            top1_correct += int(predicted_label == true_label)
+
+        binary_true_array = np.asarray(binary_true)
+        binary_scores_array = np.asarray(binary_scores)
+        auc = float(roc_auc_score(binary_true_array, binary_scores_array))
+        eer = equal_error_rate(binary_true_array, binary_scores_array)
+        fold_rows.append(
+            {
+                "fold": fold,
+                "roc_auc": auc,
+                "eer": eer,
+                "tar_at_far_1pct": tar_at_far(binary_true_array, binary_scores_array, 0.01),
+                "top1_accuracy": top1_correct / max(len(test_idx), 1),
+                "n_test": int(len(test_idx)),
+            }
+        )
+        if verbose:
+            print(
+                f"[verification] фолд {fold}/{n_splits}: "
+                f"ROC-AUC={auc:.4f}, EER={eer:.4f}, top1={top1_correct / max(len(test_idx), 1):.4f}"
+            )
+
+    fold_metrics = pd.DataFrame(fold_rows)
+    scores = pd.DataFrame(score_rows)
+    summary = {
+        column: float(fold_metrics[column].mean())
+        for column in ("roc_auc", "eer", "tar_at_far_1pct", "top1_accuracy")
+    }
+    return VerificationResult(fold_metrics=fold_metrics, scores=scores, summary=summary)
diff --git a/lab1/tests/test_cleaning.py b/lab1/tests/test_cleaning.py
new file mode 100644
index 0000000..927fc4c
--- /dev/null
+++ b/lab1/tests/test_cleaning.py
@@ -0,0 +1,33 @@
+"""Тесты нормализации текста."""
+
+from __future__ import annotations
+
+from stylometry.cleaning import (
+    clean_text,
+    count_placeholders,
+    normalize_placeholders,
+    normalize_unicode,
+    normalized_key,
+)
+
+
+def test_placeholders_are_replaced() -> None:
+    text = "Hello, <PLAYER>! Your <FARM> is nice."
+    assert normalize_placeholders(text) == "Hello, PLAYER! Your FARM is nice."
+    assert count_placeholders(text) == 2
+
+
+def test_unknown_placeholder_becomes_entity() -> None:
+    assert normalize_placeholders("Look, <UNKNOWN>!") == "Look, ENTITY!"
+
+
+def test_unicode_normalization() -> None:
+    assert normalize_unicode("Oh… really—yes") == "Oh... really-yes"
+
+
+def test_whitespace_collapse() -> None:
+    assert clean_text("  Hello\n\n  world  ") == "Hello world"
+
+
+def test_normalized_key_ignores_case_and_punctuation() -> None:
+    assert normalized_key("Oh, <PLAYER>!") == normalized_key("oh player")
diff --git a/lab1/tests/test_cv.py b/lab1/tests/test_cv.py
new file mode 100644
index 0000000..23d51fd
--- /dev/null
+++ b/lab1/tests/test_cv.py
@@ -0,0 +1,35 @@
+"""Тесты кросс-валидации на синтетических данных."""
+
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+from stylometry.cv import run_repeated_cv
+
+
+def test_repeated_cv_runs_and_has_all_predictions() -> None:
+    rng = np.random.default_rng(0)
+    n_samples = 60
+    n_classes = 3
+    X = pd.DataFrame(
+        {
+            "feature_1": rng.normal(size=n_samples),
+            "feature_2": rng.normal(size=n_samples),
+        }
+    )
+    y = np.repeat([f"c{index}" for index in range(n_classes)], 20)
+    from sklearn.linear_model import LogisticRegression
+    from sklearn.pipeline import Pipeline
+    from sklearn.preprocessing import StandardScaler
+
+    pipeline = Pipeline(
+        [("scale", StandardScaler()), ("model", LogisticRegression(max_iter=1000))]
+    )
+    result = run_repeated_cv(
+        pipeline, X, y, n_splits=3, n_repeats=2, seed=42, verbose=False
+    )
+    assert len(result.oof) == n_samples * 2
+    assert set(result.oof["y_true"]) == set(y)
+    assert result.fold_metrics["macro_f1"].between(0, 1).all()
+    assert result.labels.tolist() == ["c0", "c1", "c2"]
diff --git a/lab1/tests/test_delta.py b/lab1/tests/test_delta.py
new file mode 100644
index 0000000..c9351c5
--- /dev/null
+++ b/lab1/tests/test_delta.py
@@ -0,0 +1,41 @@
+"""Тесты дельты Барроуза."""
+
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+from stylometry.delta import DeltaClassifier, MFWVectorizer
+
+
+def _toy_frame() -> pd.DataFrame:
+    texts = [
+        "the cat sat on the mat",
+        "the cat sat on the mat again",
+        "the dog sat on the log",
+        "the dog sat on the log again",
+        "a cat and a dog",
+        "a cat and a dog again",
+    ]
+    labels = ["cat", "cat", "dog", "dog", "cat", "dog"]
+    return pd.DataFrame({"text": texts, "character": labels})
+
+
+def test_mfw_vectorizer_shape() -> None:
+    frame = _toy_frame()
+    vectorizer = MFWVectorizer(n_features=5)
+    matrix = vectorizer.fit_transform(frame)
+    assert matrix.shape == (6, 5)
+    assert np.isfinite(matrix).all()
+
+
+def test_delta_fit_predict() -> None:
+    frame = _toy_frame()
+    classifier = DeltaClassifier(n_features=5, metric="manhattan")
+    classifier.fit(frame, frame["character"].values)
+    predictions = classifier.predict(frame)
+    assert len(predictions) == len(frame)
+    assert set(predictions) <= {"cat", "dog"}
+    probabilities = classifier.predict_proba(frame)
+    assert probabilities.shape == (len(frame), 2)
+    assert np.allclose(probabilities.sum(axis=1), 1.0)
diff --git a/lab1/tests/test_features.py b/lab1/tests/test_features.py
new file mode 100644
index 0000000..b6dddea
--- /dev/null
+++ b/lab1/tests/test_features.py
@@ -0,0 +1,38 @@
+"""Тесты ручных стилометрических признаков."""
+
+from __future__ import annotations
+
+import numpy as np
+
+from stylometry.datasets import load_feature_frame
+from stylometry.features import StylometricVectorizer, feature_dictionary
+
+
+def test_feature_matrix_shape_and_finiteness() -> None:
+    frame = load_feature_frame(verbose=False).head(20)
+    vectorizer = StylometricVectorizer()
+    matrix = vectorizer.fit_transform(frame)
+    names = vectorizer.get_feature_names_out()
+    assert matrix.shape == (20, len(names))
+    assert np.isfinite(matrix).all()
+    assert len(set(names)) == len(names)
+
+
+def test_feature_dictionary() -> None:
+    frame = load_feature_frame(verbose=False).head(2)
+    vectorizer = StylometricVectorizer()
+    vectorizer.fit(frame)
+    dictionary = feature_dictionary(vectorizer.get_feature_names_out())
+    assert len(dictionary) == len(vectorizer.get_feature_names_out())
+    assert {"feature", "group", "description", "status"} <= set(dictionary.columns)
+
+
+def test_group_subset() -> None:
+    frame = load_feature_frame(verbose=False).head(5)
+    vectorizer = StylometricVectorizer(groups=["length", "punctuation"])
+    matrix = vectorizer.fit_transform(frame)
+    assert matrix.shape[0] == 5
+    assert all(
+        name.startswith("len_") or name.startswith("punct_")
+        for name in vectorizer.get_feature_names_out()
+    )
diff --git a/lab1/tests/test_io.py b/lab1/tests/test_io.py
new file mode 100644
index 0000000..e01781e
--- /dev/null
+++ b/lab1/tests/test_io.py
@@ -0,0 +1,52 @@
+"""Тесты чтения и валидации корпуса."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pandas as pd
+
+from stylometry.io import (
+    DEFAULT_DATA_PATH,
+    compare_with_stats,
+    corpus_fingerprint,
+    read_corpus,
+    read_stats,
+    validate_corpus,
+)
+
+
+def test_read_corpus_shape() -> None:
+    frame = read_corpus()
+    assert len(frame) == 2005
+    assert frame["character"].nunique() == 34
+    assert {"line_id", "text", "character", "text_raw", "placeholder_count"} <= set(frame.columns)
+    assert frame["text"].str.len().gt(0).all()
+
+
+def test_corpus_is_cp1251() -> None:
+    raw = DEFAULT_DATA_PATH.read_bytes()
+    assert raw.decode("cp1251").startswith("Реплика;Персонаж")
+
+
+def test_validate_corpus() -> None:
+    frame = read_corpus()
+    stats = validate_corpus(frame)
+    assert stats["n_lines"] == 2005
+    assert stats["n_characters"] == 34
+    assert stats["n_duplicate_texts"] == 0
+    assert stats["n_words"] > 30_000
+
+
+def test_compare_with_stats() -> None:
+    frame = read_corpus()
+    comparison = compare_with_stats(frame, read_stats())
+    assert not comparison.empty
+    assert (comparison["difference"] == 0).all()
+
+
+def test_fingerprint_is_stable() -> None:
+    first = corpus_fingerprint()
+    second = corpus_fingerprint()
+    assert first == second
+    assert len(first) == 64
diff --git a/lab1/tests/test_lsa.py b/lab1/tests/test_lsa.py
new file mode 100644
index 0000000..77e9112
--- /dev/null
+++ b/lab1/tests/test_lsa.py
@@ -0,0 +1,45 @@
+"""Тесты TF-IDF и LSA."""
+
+from __future__ import annotations
+
+import numpy as np
+
+from stylometry.datasets import load_feature_frame
+from stylometry.lsa import (
+    make_char_ngram_vectorizer,
+    make_lsa_vectorizer,
+    make_pos_ngram_vectorizer,
+    make_text_vectorizer,
+)
+
+
+def test_lsa_word_shape_and_variance() -> None:
+    frame = load_feature_frame(verbose=False).head(100)
+    vectorizer = make_lsa_vectorizer("word", n_components=10)
+    matrix = vectorizer.fit_transform(frame)
+    assert matrix.shape == (100, 10)
+    assert np.isfinite(matrix).all()
+    assert 0 <= vectorizer.cumulative_explained_variance_[-1] <= 1.0
+    assert len(vectorizer.top_terms(5)) == 10
+
+
+def test_lsa_pos_and_char() -> None:
+    frame = load_feature_frame(verbose=False).head(100)
+    for variant in ("pos", "char"):
+        vectorizer = make_lsa_vectorizer(variant, n_components=5)
+        matrix = vectorizer.fit_transform(frame)
+        assert matrix.shape[0] == 100
+        assert matrix.shape[1] == 5
+        assert np.isfinite(matrix).all()
+
+
+def test_text_vectorizers() -> None:
+    frame = load_feature_frame(verbose=False).head(50)
+    for vectorizer in (
+        make_text_vectorizer("word"),
+        make_pos_ngram_vectorizer(),
+        make_char_ngram_vectorizer(),
+    ):
+        matrix = vectorizer.fit_transform(frame)
+        assert matrix.shape[0] == 50
+        assert matrix.shape[1] > 0
diff --git a/lab1/tests/test_metadata.py b/lab1/tests/test_metadata.py
new file mode 100644
index 0000000..c95a048
--- /dev/null
+++ b/lab1/tests/test_metadata.py
@@ -0,0 +1,36 @@
+"""Тесты анализа по метаданным персонажей."""
+
+from __future__ import annotations
+
+import pandas as pd
+
+from stylometry.metadata import (
+    group_confusions,
+    load_character_metadata,
+    subgroup_report,
+)
+
+
+def test_metadata_table() -> None:
+    metadata = load_character_metadata()
+    assert len(metadata) == 34
+    assert {"character", "gender", "age_group", "species", "role"} <= set(metadata.columns)
+
+
+def test_subgroup_report() -> None:
+    metadata = load_character_metadata()
+    y_true = ["Abigail", "Abigail", "Alex", "Alex"]
+    y_pred = ["Abigail", "Alex", "Alex", "Alex"]
+    labels = ["Abigail", "Alex"]
+    report = subgroup_report(y_true, y_pred, labels, metadata, "gender")
+    assert {"group", "weighted_f1", "n_samples"} <= set(report.columns)
+    assert report["n_samples"].sum() == 4
+
+
+def test_group_confusions() -> None:
+    metadata = load_character_metadata()
+    y_true = ["Abigail", "Abigail", "Alex", "Alex"]
+    y_pred = ["Abigail", "Alex", "Alex", "Abigail"]
+    table = group_confusions(y_true, y_pred, metadata, "gender")
+    assert not table.empty
+    assert {"true_group", "pred_group", "count"} <= set(table.columns)
diff --git a/lab1/tests/test_metrics.py b/lab1/tests/test_metrics.py
new file mode 100644
index 0000000..7d991c5
--- /dev/null
+++ b/lab1/tests/test_metrics.py
@@ -0,0 +1,72 @@
+"""Тесты метрик и статистических критериев."""
+
+from __future__ import annotations
+
+import numpy as np
+import pandas as pd
+
+from stylometry.evaluation import (
+    bootstrap_ci,
+    classification_metrics,
+    confusion_matrix_frame,
+    per_class_report,
+)
+from stylometry.stats import (
+    bootstrap_difference,
+    holm_bonferroni,
+    mcnemar_exact,
+    wilcoxon_signed_rank,
+)
+
+
+def test_classification_metrics_perfect() -> None:
+    y_true = np.array(["a", "b", "c", "a", "b", "c"])
+    y_pred = y_true.copy()
+    metrics = classification_metrics(y_true, y_pred)
+    assert metrics["accuracy"] == 1.0
+    assert metrics["macro_f1"] == 1.0
+    assert metrics["balanced_accuracy"] == 1.0
+
+
+def test_confusion_matrix_and_report() -> None:
+    y_true = np.array(["a", "a", "b", "b"])
+    y_pred = np.array(["a", "b", "b", "b"])
+    matrix = confusion_matrix_frame(y_true, y_pred, ["a", "b"])
+    assert matrix.loc["a", "a"] == 1
+    report = per_class_report(y_true, y_pred, ["a", "b"])
+    assert {"precision", "recall", "f1", "support"} <= set(report.columns)
+
+
+def test_bootstrap_ci() -> None:
+    y_true = np.array(["a", "b", "a", "b", "a", "b"])
+    y_pred = np.array(["a", "b", "a", "a", "a", "b"])
+    interval = bootstrap_ci(y_true, y_pred, ["a", "b"], n_boot=50, seed=1)
+    assert interval["lower"] <= interval["upper"]
+
+
+def test_mcnemar() -> None:
+    y_true = np.array([1, 1, 0, 0, 1, 0])
+    pred_a = np.array([1, 1, 0, 0, 1, 0])
+    pred_b = np.array([1, 0, 0, 1, 1, 0])
+    result = mcnemar_exact(y_true, pred_a, pred_b)
+    assert result["b"] >= 0 and result["c"] >= 0
+    assert 0 <= result["p_value"] <= 1
+
+
+def test_holm() -> None:
+    adjusted = holm_bonferroni([0.01, 0.02, 0.03])
+    assert len(adjusted) == 3
+    assert all(0 <= value <= 1 for value in adjusted)
+
+
+def test_wilcoxon() -> None:
+    result = wilcoxon_signed_rank([0.5, 0.6, 0.7], [0.4, 0.6, 0.5])
+    assert 0 <= result["p_value"] <= 1
+
+
+def test_bootstrap_difference() -> None:
+    y_true = np.array([0, 1, 2, 0, 1, 2])
+    pred_a = np.array([0, 1, 2, 0, 1, 2])
+    pred_b = np.array([0, 1, 1, 0, 1, 1])
+    result = bootstrap_difference(y_true, pred_a, pred_b, labels=[0, 1, 2], n_boot=50)
+    assert result["lower"] <= result["upper"]
diff --git a/lab1/tests/test_tokenization.py b/lab1/tests/test_tokenization.py
new file mode 100644
index 0000000..548c1ef
--- /dev/null
+++ b/lab1/tests/test_tokenization.py
@@ -0,0 +1,40 @@
+"""Тесты NLTK-токенизации и частеречной разметки."""
+
+from __future__ import annotations
+
+from stylometry.tokenization import Tokenizer, penn_to_upos
+
+
+def test_penn_to_upos() -> None:
+    assert penn_to_upos("the", "DT") == "DET"
+    assert penn_to_upos("run", "VB") == "VERB"
+    assert penn_to_upos("is", "VBZ") == "AUX"
+    assert penn_to_upos("because", "IN") == "SCONJ"
+    assert penn_to_upos("quickly", "RB") == "ADV"
+
+
+def test_analyze_line() -> None:
+    tokenizer = Tokenizer(use_nltk=True)
+    analyzed = tokenizer.analyze("Oh... hi. I don't know, Mr. Qi.")
+    assert analyzed.tokens
+    assert len(analyzed.tokens) == len(analyzed.upos) == len(analyzed.ptb)
+    assert analyzed.sentences
+    assert analyzed.pos_text
+    assert analyzed.fw_text
+    assert analyzed.content_text
+
+
+def test_placeholders_are_tagged_as_words() -> None:
+    tokenizer = Tokenizer(use_nltk=True)
+    analyzed = tokenizer.analyze("Hello, PLAYER! You are nice.")
+    assert "PLAYER" in analyzed.tokens
+    assert len(analyzed.upos) == len(analyzed.tokens)
+
+
+def test_fallback_regex_tokenizer() -> None:
+    tokenizer = Tokenizer(use_nltk=False)
+    analyzed = tokenizer.analyze("Hello, world! It's fine.")
+    assert analyzed.tokens == ["Hello", ",", "world", "!", "It's", "fine", "."]
+    assert len(analyzed.upos) == len(analyzed.tokens)
+    assert analyzed.upos[0] == "X"
+    assert "PUNCT" in analyzed.upos
diff --git a/lab1/О первой практической работе (ААТ, ИВТ).pdf b/lab1/О первой практической работе (ААТ, ИВТ).pdf
new file mode 100644
index 0000000..3f43bb1
Binary files /dev/null and b/lab1/О первой практической работе (ААТ, ИВТ).pdf differ