Upload Task 3 VLM benchmark dataset from Colab

Generated from Task 3 Google Sheet URLs and adapted for VLM A/B generation.

ARTICLE_IMAGE_SOURCES.md

@@ -1,11 +1,11 @@
 # Источники статей для изображений
 
-Файл создан автоматически: `2026-05-28T19:45:17Z`.
+Файл создан автоматически: `2026-06-03T07:06:18Z`.
 
 Назначение файла — зафиксировать ссылки на статьи, из которых взяты изображения/отрендеренные PDF-страницы в `data/task3_vlm_generation.jsonl`. Это нужно для соблюдения правил цитирования источников.
 
-Всего image-записей: **1440**.
-Уникальных статей/идентификаторов: **192**.
+Всего image-записей: **1496**.
+Уникальных статей/идентификаторов: **197**.
 
 ## 978-1-939133-44-1
 
@@ -97,6 +97,28 @@
 | `task3:ca35bd9ad6:case-009` | `CASE-009` | `assets/images/ca35bd9ad6/case-009/explicit_03_page_010.png` | p.10, Fig. 11 | figure |
 | `task3:ca35bd9ad6:case-009` | `CASE-009` | `assets/images/ca35bd9ad6/case-009/explicit_04_page_010.png` | p.10, Fig. 11 | figure |
 
+## https://arxiv.org/abs/1605.01735
+
+- Ссылка на статью: https://arxiv.org/abs/1605.01735
+- Название: Machine learning phases of matter
+- Идентификаторы: arxiv:1605.01735
+- Использовано изображений/страниц: 12
+
+| sample_id | case_id | image | page_hint | evidence_kind |
+|---|---|---|---|---|
+| `task3:task3-ab-physics-ml-001:case_01` | `case_01` | `assets/images/task3-ab-physics-ml-001/case_01/explicit_01_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_01` | `case_01` | `assets/images/task3-ab-physics-ml-001/case_01/explicit_02_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_01` | `case_01` | `assets/images/task3-ab-physics-ml-001/case_01/explicit_03_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_01` | `case_01` | `assets/images/task3-ab-physics-ml-001/case_01/explicit_04_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_02` | `case_02` | `assets/images/task3-ab-physics-ml-001/case_02/explicit_01_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_02` | `case_02` | `assets/images/task3-ab-physics-ml-001/case_02/explicit_02_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_02` | `case_02` | `assets/images/task3-ab-physics-ml-001/case_02/explicit_03_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_02` | `case_02` | `assets/images/task3-ab-physics-ml-001/case_02/explicit_04_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_03` | `case_03` | `assets/images/task3-ab-physics-ml-001/case_03/explicit_01_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_03` | `case_03` | `assets/images/task3-ab-physics-ml-001/case_03/explicit_02_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_03` | `case_03` | `assets/images/task3-ab-physics-ml-001/case_03/explicit_03_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_03` | `case_03` | `assets/images/task3-ab-physics-ml-001/case_03/explicit_04_page_003.png` | Fig. 3 | figure |
+
 ## https://arxiv.org/abs/1704.07994
 
 - Ссылка на статью: https://arxiv.org/abs/1704.07994
@@ -313,6 +335,28 @@
 | `task3:cherbaev__iris_retina_modern_ab:case-010` | `CASE-010` | `assets/images/cherbaev__iris_retina_modern_ab/case-010/explicit_03_page_000.png` | Таблица или график match scores по 6 сессиям за 6.5 лет, 209 субъектов (4–11 лет на enrollment) | figure_or_table |
 | `task3:cherbaev__iris_retina_modern_ab:case-010` | `CASE-010` | `assets/images/cherbaev__iris_retina_modern_ab/case-010/explicit_04_page_000.png` | Таблица или график match scores по 6 сессиям за 6.5 лет, 209 субъектов (4–11 лет на enrollment) | figure_or_table |
 
+## https://arxiv.org/abs/2305.02402
+
+- Ссылка на статью: https://arxiv.org/abs/2305.02402
+- Название: Normalizing flows for lattice gauge theory in arbitrary space-time dimension
+- Идентификаторы: arxiv:2305.02402
+- Использовано изображений/страниц: 12
+
+| sample_id | case_id | image | page_hint | evidence_kind |
+|---|---|---|---|---|
+| `task3:task3-ab-physics-ml-001:case_04` | `case_04` | `assets/images/task3-ab-physics-ml-001/case_04/explicit_01_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_04` | `case_04` | `assets/images/task3-ab-physics-ml-001/case_04/explicit_02_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_04` | `case_04` | `assets/images/task3-ab-physics-ml-001/case_04/explicit_03_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_04` | `case_04` | `assets/images/task3-ab-physics-ml-001/case_04/explicit_04_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_05` | `case_05` | `assets/images/task3-ab-physics-ml-001/case_05/explicit_01_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_05` | `case_05` | `assets/images/task3-ab-physics-ml-001/case_05/explicit_02_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_05` | `case_05` | `assets/images/task3-ab-physics-ml-001/case_05/explicit_03_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_05` | `case_05` | `assets/images/task3-ab-physics-ml-001/case_05/explicit_04_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_09` | `case_09` | `assets/images/task3-ab-physics-ml-001/case_09/explicit_01_page_006.png` | Fig. 6 | figure |
+| `task3:task3-ab-physics-ml-001:case_09` | `case_09` | `assets/images/task3-ab-physics-ml-001/case_09/explicit_02_page_006.png` | Fig. 6 | figure |
+| `task3:task3-ab-physics-ml-001:case_09` | `case_09` | `assets/images/task3-ab-physics-ml-001/case_09/explicit_03_page_016.png` | Fig. 6 | figure |
+| `task3:task3-ab-physics-ml-001:case_09` | `case_09` | `assets/images/task3-ab-physics-ml-001/case_09/explicit_04_page_016.png` | Fig. 6 | figure |
+
 ## https://arxiv.org/abs/2305.16432
 
 - Ссылка на статью: https://arxiv.org/abs/2305.16432
@@ -651,6 +695,28 @@
 | `task3:gorbunov-ivan-yurievich__0d769c50eb:case-012` | `CASE-012` | `assets/images/gorbunov-ivan-yurievich__0d769c50eb/case-012/explicit_03_page_004.png` | Fig.4 | figure |
 | `task3:gorbunov-ivan-yurievich__0d769c50eb:case-012` | `CASE-012` | `assets/images/gorbunov-ivan-yurievich__0d769c50eb/case-012/explicit_04_page_004.png` | Fig.4 | figure |
 
+## https://arxiv.org/abs/2405.16769
+
+- Ссылка на статью: https://arxiv.org/abs/2405.16769
+- Название: Learning phase transitions by siamese neural network
+- Идентификаторы: arxiv:2405.16769
+- Использовано изображений/страниц: 12
+
+| sample_id | case_id | image | page_hint | evidence_kind |
+|---|---|---|---|---|
+| `task3:task3-ab-physics-ml-001:case_07` | `case_07` | `assets/images/task3-ab-physics-ml-001/case_07/explicit_01_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_07` | `case_07` | `assets/images/task3-ab-physics-ml-001/case_07/explicit_02_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_07` | `case_07` | `assets/images/task3-ab-physics-ml-001/case_07/explicit_03_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_07` | `case_07` | `assets/images/task3-ab-physics-ml-001/case_07/explicit_04_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_08` | `case_08` | `assets/images/task3-ab-physics-ml-001/case_08/explicit_01_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_08` | `case_08` | `assets/images/task3-ab-physics-ml-001/case_08/explicit_02_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_08` | `case_08` | `assets/images/task3-ab-physics-ml-001/case_08/explicit_03_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_08` | `case_08` | `assets/images/task3-ab-physics-ml-001/case_08/explicit_04_page_004.png` | Fig. 4 | figure |
+| `task3:task3-ab-physics-ml-001:case_12` | `case_12` | `assets/images/task3-ab-physics-ml-001/case_12/explicit_01_page_005.png` | Fig. 5 | figure |
+| `task3:task3-ab-physics-ml-001:case_12` | `case_12` | `assets/images/task3-ab-physics-ml-001/case_12/explicit_02_page_005.png` | Fig. 5 | figure |
+| `task3:task3-ab-physics-ml-001:case_12` | `case_12` | `assets/images/task3-ab-physics-ml-001/case_12/explicit_03_page_005.png` | Fig. 5 | figure |
+| `task3:task3-ab-physics-ml-001:case_12` | `case_12` | `assets/images/task3-ab-physics-ml-001/case_12/explicit_04_page_005.png` | Fig. 5 | figure |
+
 ## https://arxiv.org/abs/2407.03037
 
 - Ссылка на статью: https://arxiv.org/abs/2407.03037
@@ -711,6 +777,28 @@
 | `task3:vasin_artem_aleksandrovich__0cf7f9ceca:case-002` | `CASE-002` | `assets/images/vasin_artem_aleksandrovich__0cf7f9ceca/case-002/explicit_03_page_002.png` | fig. 2, 3, 6, Tab. 2 | figure_or_table |
 | `task3:vasin_artem_aleksandrovich__0cf7f9ceca:case-002` | `CASE-002` | `assets/images/vasin_artem_aleksandrovich__0cf7f9ceca/case-002/explicit_04_page_002.png` | fig. 2, 3, 6, Tab. 2 | figure_or_table |
 
+## https://arxiv.org/abs/2410.13161
+
+- Ссылка на статью: https://arxiv.org/abs/2410.13161
+- Название: Non-Perturbative Trivializing Flows for Lattice Gauge Theories
+- Идентификаторы: arxiv:2410.13161
+- Использовано изображений/страниц: 12
+
+| sample_id | case_id | image | page_hint | evidence_kind |
+|---|---|---|---|---|
+| `task3:task3-ab-physics-ml-001:case_06` | `case_06` | `assets/images/task3-ab-physics-ml-001/case_06/explicit_01_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_06` | `case_06` | `assets/images/task3-ab-physics-ml-001/case_06/explicit_02_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_06` | `case_06` | `assets/images/task3-ab-physics-ml-001/case_06/explicit_03_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_06` | `case_06` | `assets/images/task3-ab-physics-ml-001/case_06/explicit_04_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_10` | `case_10` | `assets/images/task3-ab-physics-ml-001/case_10/explicit_01_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_10` | `case_10` | `assets/images/task3-ab-physics-ml-001/case_10/explicit_02_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_10` | `case_10` | `assets/images/task3-ab-physics-ml-001/case_10/explicit_03_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_10` | `case_10` | `assets/images/task3-ab-physics-ml-001/case_10/explicit_04_page_003.png` | Fig. 3 | figure |
+| `task3:task3-ab-physics-ml-001:case_13` | `case_13` | `assets/images/task3-ab-physics-ml-001/case_13/explicit_01_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_13` | `case_13` | `assets/images/task3-ab-physics-ml-001/case_13/explicit_02_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_13` | `case_13` | `assets/images/task3-ab-physics-ml-001/case_13/explicit_03_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_13` | `case_13` | `assets/images/task3-ab-physics-ml-001/case_13/explicit_04_page_001.png` | Fig. 1 | figure |
+
 ## https://arxiv.org/abs/2410.17881
 
 - Ссылка на статью: https://arxiv.org/abs/2410.17881
@@ -809,6 +897,24 @@
 | `task3:vasin_artem_aleksandrovich__0cf7f9ceca:case-009` | `CASE-009` | `assets/images/vasin_artem_aleksandrovich__0cf7f9ceca/case-009/explicit_03_page_001.png` | Fig. 1, 2 | figure |
 | `task3:vasin_artem_aleksandrovich__0cf7f9ceca:case-009` | `CASE-009` | `assets/images/vasin_artem_aleksandrovich__0cf7f9ceca/case-009/explicit_04_page_001.png` | Fig. 1, 2 | figure |
 
+## https://arxiv.org/abs/2501.05547
+
+- Ссылка на статью: https://arxiv.org/abs/2501.05547
+- Название: Deep learning of phase transitions with minimal examples
+- Идентификаторы: arxiv:2501.05547
+- Использовано изображений/страниц: 8
+
+| sample_id | case_id | image | page_hint | evidence_kind |
+|---|---|---|---|---|
+| `task3:task3-ab-physics-ml-001:case_11` | `case_11` | `assets/images/task3-ab-physics-ml-001/case_11/explicit_01_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_11` | `case_11` | `assets/images/task3-ab-physics-ml-001/case_11/explicit_02_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_11` | `case_11` | `assets/images/task3-ab-physics-ml-001/case_11/explicit_03_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_11` | `case_11` | `assets/images/task3-ab-physics-ml-001/case_11/explicit_04_page_002.png` | Fig. 2 | figure |
+| `task3:task3-ab-physics-ml-001:case_14` | `case_14` | `assets/images/task3-ab-physics-ml-001/case_14/explicit_01_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_14` | `case_14` | `assets/images/task3-ab-physics-ml-001/case_14/explicit_02_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_14` | `case_14` | `assets/images/task3-ab-physics-ml-001/case_14/explicit_03_page_001.png` | Fig. 1 | figure |
+| `task3:task3-ab-physics-ml-001:case_14` | `case_14` | `assets/images/task3-ab-physics-ml-001/case_14/explicit_04_page_001.png` | Fig. 1 | figure |
+
 ## https://arxiv.org/abs/2501.06312
 
 - Ссылка на статью: https://arxiv.org/abs/2501.06312

README.md

@@ -16,7 +16,7 @@ tags:
 
 Этот датасет содержит Task 3 кейсы для генерации ответа одной vision-language моделью. Исходные экспертные A/B prompts преобразуются в модельную задачу: модель должна извлечь факты из текста, страниц PDF, рисунков или таблиц и вернуть структурированный JSON-ответ. Экспертные rationale и ожидаемые ошибки вынесены в отдельные metadata-файлы, чтобы не протекать в prompt оцениваемой модели.
 
-Дата генерации README: `2026-05-28T19:45:17Z`.
+Дата генерации README: `2026-06-03T07:06:18Z`.
 
 ## Главный датасет: `data/task3_vlm_generation.jsonl`
 
@@ -31,7 +31,7 @@ tags:
 - `generation_target_schema` — ожидаемая структура JSON-ответа;
 - `review_metadata` — диагностические поля для эксперта; их не следует передавать оцениваемой VLM.
 
-Строк в текущей сборке: **372**. Строк с изображениями: **372**. Всего ссылок на изображения: **1488**.
+Строк в текущей сборке: **386**. Строк с изображениями: **386**. Всего ссылок на изображения: **1544**.
 
 ## Дополнительные файлы
 
@@ -39,7 +39,7 @@ tags:
 
 Плоская таблица кейсов для аудита. Здесь удобно проверять `case_id`, `paper_title`, `paper_id`, `creator_prompt`, `model_task_prompt`, `review_focus` и enabled/primary flags без раскрытия полной message-структуры.
 
-Строк: **413**.
+Строк: **427**.
 
 ### `data/task3_cases_summary.csv`
 
@@ -65,13 +65,13 @@ CSV-сводка по строкам генерационного датасет
 
 ## Распределение по слоям сложности
 
-- `easy_control`: 59
-- `multimodal_hard`: 220
-- `temporal_hard`: 93
+- `easy_control`: 61
+- `multimodal_hard`: 228
+- `temporal_hard`: 97
 
 ## Распределение по типам evidence
 
-- `figure`: 147
+- `figure`: 161
 - `figure_or_table`: 99
 - `formula`: 9
 - `mixed`: 23

article_image_sources.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79b85478785e8f70dfe0ebfad38aca5f8d171ba7e55bae26ae82cb66a7a801f3
-size 985764
+oid sha256:667b9f65e10de4bab5270109489009ee90e466b5a2b68042c4e16adbf0ca3e04
+size 1019688