.gitignore

# Python
__pycache__/
*.py[cod]
*$py.class
*.so
.Python
build/
develop-eggs/
dist/
downloads/
eggs/
.eggs/
lib/
lib64/
parts/
sdist/
var/
wheels/
*.egg-info/
.installed.cfg
*.egg

# Virtual environments
.venv/
venv/
ENV/

# IDE
.idea/
.vscode/
*.swp
*.swo

# Testing
.pytest_cache/
.coverage
htmlcov/
.tox/
.nox/

# MyPy
.mypy_cache/

# OS
.DS_Store
Thumbs.db

# Reference databases
# Keep compressed .sd.gz files, ignore decompressed .sd files (auto-generated)
data/reference/*.sd
!data/reference/*.sd.gz
# Ignore large SDF files (COCONUT ~4.5GB)
data/reference/*.sdf
# Ignore SQLite databases (generated via `ailsa database build`)
data/reference/*.db
# Keep compressed database for distribution
!data/reference/*.db.gz
# Ignore zipped databases (deposited on Zenodo/Figshare, not in git)
data/reference/*.db.zip

# Benchmark results (large, generated at runtime)
tests/case-benchmark/results/

# LSD runtime debris (LSD binary writes these to the cwd on every run)
solncounter
outlsd.out
# Fragment goodlist files generated by `ailsa fragment to-lsd` (kept where the agent puts them, not at repo root)
/fragment_*.lsd
# CASE/LSD run artefacts must never land at repo root (they belong in data dirs)
/*.lsd
/*.sol
/*.smi
/*.log

# Large background PDFs — kept locally for reference, deposited elsewhere (DOI/Zenodo) not in git
background/*.pdf

# Blind-UAT answer keys — keep LOCAL only, never push (public repo).
# These map CASE datasets to their ground-truth identities / UAT outcomes.
.planning/CASE-DATASET-IDENTITIES.md
.planning/CASE-UAT-LOG.md

# Scratch notes — working files, and they routinely quote compound identities
# (InChIKey/SMILES) for datasets that are still to be run blind. Never tracked.
scratchpad/

# Build/verification scratch dir (e.g. probe `uv build` artifacts) — never tracked.
/SCRATCH/