# The measurement corpus IS the differentiated asset and is published:
#   records/frontier, records/regress, records/context  (~5 MB, 159 files)
# Everything else under records/ is a 6 GB pile of LoRA adapters, training
# corpora and routing traces — derivable, huge, and not what anyone needs.
records/*
!records/frontier/
!records/regress/
!records/context/
!records/quantladder/
!records/backend/
*.jsonl.gz
*.safetensors
*.npy

# scratch
*.log
__pycache__/
*.py[cod]
.venv/
venv/
.DS_Store
.obsidian/
.pytest_cache/
*.egg-info/
dist/
build/
