__pycache__/
*.pyc
.env
.env.local
.venv/
*.egg-info/
dist/
build/
.DS_Store
.mypy_cache/
.pytest_cache/
.ruff_cache/
data/raw/
data/processed/
*.db
*.egg-info/
.coverage
.secrets.baseline
CLAUDE.md.backup
checkpoints/

# Sprint 60 Track B: any tempfile orphans (sec M-3 belt-and-braces;
# tempdir is OS-managed but defence-in-depth against accidental REPO_ROOT writes)
track_b_*.tmp.json
track_b_audit_partial_*.json
track_b_acl_*.json

# ML model artefacts — large binaries kept in GCS, not git. Text metadata
# (results.json, VERIFIED.md) stays tracked so metrics are reproducible.
models/**/*.pt
models/**/*.ckpt
models/**/*.bin
models/**/*.safetensors
models/**/training.log

lib/
!services/acc/frontend/lib/
!site/src/lib/
node_modules/
data/reports/elliptic-plus-plus/
data/reports/*.pdf
# Sprint 46 GNN penultimate embeddings — 554 MB, rebuilt by
# scripts/build_similar_embeddings.py against the checkpoint + graph
# tensors; never belongs in git.
data/embeddings/

# Downloaded graph artefacts from GCS — GCS is source of truth, scripts pull
# on demand. addr_to_idx.json is 50+ MB, node_features/edge_index are .pt.
data/graphs/

# Joint entity triplet artefacts (B.1 v5 ≥100 MiB). Canonical copies live on
# GCS at gs://aidecisions-data/entity/ — reconstruct via scripts/entity.
data/entity_triplets_*.json

# Unified labels artefacts (60-80 MiB each version). Canonical copies live
# on GCS at gs://aidecisions-data/labels/ — pulled on demand by ingestion
# scripts. Manifest with version → GCS path + sha256 in
# data/unified_labels.manifest.json (tracked).
# EXCEPTION (60.AMRO S1 deploy 2026-06-10): the *_current.json alias is baked
# into the prod Docker image so /v1/screen returns real entity category/tier
# (the scored-not-stub deploy). Tech-debt: revisit S3-pull-at-boot to drop the
# 80 MiB from git history. Other versioned dumps stay GCS-only.
data/unified_labels*.json
!data/unified_labels.manifest.json
!data/unified_labels_current.json

# OFAC / OpenSanctions / crypto-rekt scraped HTML + parsed JSON dumps.
# Many GiB each, raw scrapes regenerated by scripts/scrape_*.py.
data/labels_raw/

# Sprint 60 Track E.1.7 VASP directory — canonical copies live on
# GCS at gs://aidecisions-data/labels/. Manifest with sha256 + GCS
# path tracked at data/vasp_directory_v1.manifest.json.
data/vasp_directory_v1.parquet
data/vasp_directory_v1.json
!data/vasp_directory_v1.manifest.json
data/vasp_directory_v2.parquet
data/vasp_directory_v2.json
!data/vasp_directory_v2.manifest.json
# EXCEPTION (60.AMRO S1 deploy 2026-06-10): the v2 directory JSON is baked into
# the prod image so /v1/screen returns real VASP-licence matches (~940 KiB).
!data/vasp_directory_v2.json

# Outreach: track only .md sources + .css. Rendered html, pdf deliverables,
# and binary editor sources (.docx, .ai.pdf) stay local — regenerable from md
# or owned in external editors (Drive, Illustrator).
docs/outreach/*.html
docs/outreach/*.pdf
docs/outreach/*.docx

# Local IDE state
.vscode/
.claude/scheduled_tasks.lock

# Repo-root screenshots and ad-hoc images (not deliverables)
/H100-*.png
/maket.png
/modalka.png

# Brand assets — kept on Google Drive / Figma, repo carries only what site needs
assets/brand/*.mov
assets/brand/*.mp4
assets/brand/*.svg
config/assets/

# Site public videos and renders — large binaries, source in Drive
site/public/Logomotion-*.mp4
site/public/logomotion*.mp4
site/public/logomotion.mov
site/public/Logomotion*.mov
site/public/ai-trends-2026.pdf

# next-sitemap output (auto-generated on build)
site/public/sitemap.xml
site/public/robots.txt

# Repo-root lock-file noise (next-sitemap pulled into root by mistake)
/package-lock.json
/package.json

# Pytorch Lightning logs and screenshot artefacts
lightning_logs/
test-results/
logs/

# nvidia/ — outreach pitch deck working dir, decks owned in PowerPoint
nvidia/
publications/aidecisions_NVIDIA.pdf

# Design references — owned in Figma, local mirror only
design/

# Journal service binary artefacts (xlsx, wpress backups)
services/journal/*.xlsx
services/journal/*.wpress
services/journal/www/

# ACC test-input spreadsheets — large binaries with sample data
services/acc/data/test-inputs/
services/acc/Bank.xlsx

# Top-level research notes scoped out of repo (Notion / Drive originals)
/About_Graphify.md
/Google_Claude.md

# Crypto crime report PDFs at repo root
/the-2026-*.pdf

# ML telemetry JSONs that exceed 1MB or contain transient sweep state.
# Track only summary results.json + checkpoints metadata in models/<run>/.
models/abm_v2/ppo-training-*/cache/
models/abm_v2/ppo-training-*/policy.pt
models/abm_v2/ppo-training-*/ppo_training.log
models/abm_v2/ppo-run*/telemetry.json
models/abm_v2/ppo-training-*/telemetry.json

# Empty / placeholder data files (regenerated by scripts)
data/walletexplorer_labels.json

# S4 taint generated artefacts (live on GCS, not committed)
scripts/gcp/sprint60-amro-s4/provenance_index_v1*.json
scripts/gcp/sprint60-amro-s4/_seeds_eth.ndjson

# bridge interactions (generated, on GCS)
data/bridges/cross_chain_interactions_v1.json

# founder review sheets (global rule: <project>/reviews/, gitignored)
reviews/
