Nomearod/agentbench

1__pycache__/2*.py[cod]3*.egg-info/4dist/5build/6.eggs/7*.egg8.cache/9.mypy_cache/10.pytest_cache/11.ruff_cache/12*.faiss13*.pkl14.env15.env.*16.env*17.venv/18venv/19.worktrees/20*.db21 22# Runtime audit / telemetry logs — contain hashed IPs, raw prompts,23# security verdicts. Never commit these.24logs/25*.jsonl26 27# Evidence-bearing measurement artifacts referenced from DECISIONS.md.28# Narrow exception to the *.jsonl ignore above. Add new measurement files29# explicitly here so the audit-trail intent stays opt-in.30!measurements/*.jsonl31# Calibration jury/permute sidecars (per-member detail for κ ablation table).32!results/*.jsonl33 34# Opaque binary artifacts — no PDFs in the repo today, and any that35# appear here are almost always local reference material (downloaded36# papers, vendor docs) that should not be committed. If a PDF ever37# needs to be tracked for real, add it with an explicit force-add and38# a targeted gitignore exception next to it.39*.pdf40docs/DESIGN.md41# Internal planning and design docs (not published)42docs/plans/43docs/superpowers/44# Generated statistics report (make evaluate-stats). Regenerable on demand45# from results/long/; the committed artifacts are the test goldens under46# tests/stats/fixtures/, not this live output.47docs/_generated/48terraform.tfvars49.terraform/50*.tfstate51*.tfstate.backup52.DS_Store53 54# Epoch-runner outputs from `make epochs` (PAID, human-run): raw per-config55# outputs and run-id envelope dirs. Real-run results must not be accidentally56# committed or git-clean'd; convert them to results/long/ via the adapter.57results/epochs/58 59# Claude Code local state: machine-local settings and transient runtime60# locks. settings.local.json is per-machine; a shareable settings.json,61# if ever added, stays committable. The scheduler lock is runtime state.62# Neither belongs in version control.63.claude/settings.local.json64.claude/*.lock65.venv-phase2/66