.gitignore

# Python
__pycache__/
*.py[cod]
*$py.class
*.egg-info/
.eggs/
dist/
build/
*.so

# Virtual envs
.venv/
venv/
env/
ENV/
.env
.env.local
.env.*.local

# uv / poetry
.uv-cache/
poetry.lock.bak

# Testing
.pytest_cache/
.tox/
.coverage
.coverage.*
htmlcov/
coverage.xml
.hypothesis/

# Type checking
.mypy_cache/
.ruff_cache/
.pyre/
.pytype/

# IDE
.idea/
.vscode/
*.swp
*.swo
.DS_Store
Thumbs.db

# Project-specific
# BIRD dump + Chinook DB. Only the SQLite engines themselves are
# committed (the runtime needs them); eval-only artefacts (gold SQL,
# database_description CSVs, the MINIDEV_*sql/ siblings, and the
# original minidev.zip) stay ignored. Three BIRD DBs (>= 250 MB each:
# card_games, codebase_community, european_football_2) bust GitHub's
# 100 MB hard limit and stay ignored too — they only matter for
# local eval.
#
# Pattern: ignore data/** by default and re-include only the SQLite
# engines we want shipped. git doesn't allow re-include if the
# parent directory is itself excluded — see man gitignore(5) — so
# we have to walk every parent.
data/*
data/**/*
# Walk parents we need to traverse
!data/chinook/
!data/bird_mini_dev/
!data/bird_mini_dev/MINIDEV/
!data/bird_mini_dev/MINIDEV/dev_databases/
!data/bird_mini_dev/MINIDEV/dev_databases/california_schools/
!data/bird_mini_dev/MINIDEV/dev_databases/debit_card_specializing/
!data/bird_mini_dev/MINIDEV/dev_databases/financial/
!data/bird_mini_dev/MINIDEV/dev_databases/formula_1/
!data/bird_mini_dev/MINIDEV/dev_databases/student_club/
!data/bird_mini_dev/MINIDEV/dev_databases/superhero/
!data/bird_mini_dev/MINIDEV/dev_databases/thrombosis_prediction/
!data/bird_mini_dev/MINIDEV/dev_databases/toxicology/
# Re-include only the SQLite engines themselves
!data/chinook/Chinook.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/california_schools/california_schools.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/debit_card_specializing/debit_card_specializing.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/financial/financial.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/formula_1/formula_1.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/student_club/student_club.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/superhero/superhero.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/thrombosis_prediction/thrombosis_prediction.sqlite
!data/bird_mini_dev/MINIDEV/dev_databases/toxicology/toxicology.sqlite
# BIRD train parquet (9 428 Q→SQL pairs, 2.3 MB) — used by
# `scripts/build_fewshot_index.py` to populate the cross-db fewshot
# pool. Source: huggingface.co/datasets/xu3kev/BIRD-SQL-data-train.
!data/bird_train.parquet
# Arcwise-Plat corrected-gold artefacts (Jin et al., CIDR/VLDB 2026,
# arXiv:2601.08778). ~300 KB each. Used by scripts/rescore_arcwise.py to
# stress-test our EA on annotation-error-corrected BIRD.
!data/arcwise_plat_sql_only.json
!data/arcwise_plat_full.json
# Chroma vector store. Committing the prebuilt index so the deployed
# app can serve queries without first re-embedding via the Mistral
# API. Local rebuilds via build_index.py overwrite this directory.
!chroma_data/
!chroma_data/**
chroma_data.*/
# legacy diskcache root — superseded by .cache/llm/
.diskcache/
# diskcache root: nl_sql.llm.cache writes gen/ + embed/ here
.cache/
# vcr.py cassettes (committed are in tests/cassettes/)
.cassettes/
.tmp/
*.log

# Reviews scratch (полные обзоры коммитятся, runtime артефакты — нет)
reviews/*.err

# Secrets — никогда не коммитить
secrets/
credentials/
*.pem
*.key
config/local.*
**/local.yml

# Docker
.docker-compose.override.yml

# Frontend (если будет Next.js)
frontend/node_modules/
frontend/.next/
frontend/out/
frontend/.env.local

# Notebooks
.ipynb_checkpoints/
*.nbconvert.ipynb

# OS
.Trashes
ehthumbs.db
Desktop.ini
.deploy_helper.py
.deploy_hf.py
.tmp_deploy.log