Documentation review and data consistency.

AGENTS.md

@@ -0,0 +1,51 @@
+# Repository Guidelines
+
+## Project Structure & Module Organization
+
+- `docs/`: Thesis chapters 00-07 in Spanish (UNIR structure). Edit these for narrative changes.
+- `src/`: OCR tuning code, services, notebooks, and results. Key subfolders: `raytune/`, `paddle_ocr/`, `doctr_service/`, `easyocr_service/`, `results/`, `results/correlations/`.
+- `instructions/`: UNIR template and writing rules (`plantilla_individual.htm` is the styling source of truth).
+- `thesis_output/`: Generated thesis HTML and figures (do not edit by hand).
+- Root scripts: `generate_mermaid_figures.py` (Mermaid to PNG) and `apply_content.py` (template assembly).
+- Temporary scripts go in `tem/scripts/`.
+
+## Build, Test, and Development Commands
+
+- `source .venv/bin/activate` before installing or running Python tools.
+- `npm install`: install Mermaid CLI (`node_modules/.bin/mmdc`) for figure generation.
+- `python3 generate_mermaid_figures.py`: write PNGs to `thesis_output/figures/` from `docs/*.md`.
+- `python3 apply_content.py`: generate `thesis_output/plantilla_individual.htm` from `docs/` + `instructions/`.
+- `jupyter notebook src/prepare_dataset.ipynb`: prepare OCR dataset from PDFs.
+- `jupyter notebook src/paddle_ocr_fine_tune_unir_raytune.ipynb`: run the main tuning experiment.
+- Docker tuning (GPU):
+  - `docker compose -f src/docker-compose.tuning.paddle.yml up -d paddle-ocr-gpu`
+  - `docker compose -f src/docker-compose.tuning.paddle.yml run raytune --service paddle --samples 64`
+  - `docker compose -f src/docker-compose.tuning.paddle.yml down`
+- Use `.claude/commands/word-generation.md` to regenerate the thesis output.
+
+## Coding Style & Naming Conventions
+
+- Python: PEP 8, 4-space indentation, `snake_case`.
+- Notebooks live in `src/` and should keep execution order clean when committed.
+- Documentation in `docs/` is Spanish; code comments stay in English.
+
+## Data, Documentation, and Formatting Rules
+
+- Run `.claude/commands/documentation-review.md` before editing `docs/00-07`.
+- Do not invent numbers. Every numeric claim must come from `src/results/*.csv`, `src/results/correlations/*.csv`, `docs/metrics/*.md`, or external sources listed in `docs/07_anexo_a.md`.
+- Tables and figures must use UNIR caption format: `**Tabla N.** *Título.*` / `**Figura N.** *Título.*` plus `*Fuente: ...*`.
+- Mermaid diagrams require YAML frontmatter with a quoted `title:` and UNIR theme variables.
+- Use full repository links in `*Fuente:*` lines, e.g. `https://seryus.ddns.net/unir/MastersThesis/src/branch/main/docs/metrics/metrics.md`.
+
+## Testing Guidelines
+
+- No automated tests. Validate changes by running a small tuning run and checking CSV output in `src/results/`.
+
+## Commit & Pull Request Guidelines
+
+- Commit messages are short, sentence case, and may include a tracker reference in parentheses.
+- Keep commits focused; mention generated outputs (figures, HTML) when relevant.
+
+## Agent-Specific Notes
+
+- Follow `claude.md` for thesis-specific constraints and templates.