.gitignore

# Build artifacts
.build/
podcast_scraper.egg-info/
.wily/
# setuptools/wheel output. `pip wheel .` and `python -m build` drop these at the repo
# root; they were untracked-but-not-ignored, so `git status` listed them and `git add -A`
# would have committed a build tree. The web bundles land in web/*/dist/, which the
# leading-slash rules below do not cover, hence the unanchored `dist/`.
build/
dist/
*.egg-info/

# LanceDB search index + topic_clusters.json built at CI step 0 for the synthetic
# validation corpus (GH #774). Binary, embedding-model-hash-keyed — would
# churn on every model bump and bloat the repo. Built via
# ``make build-validation-index``.
tests/fixtures/viewer-validation-corpus/search/
tests/fixtures/viewer-validation-corpus/**/search/

# v2.6.1 multi-run corpus fixture (#822) — auto-generated by the
# scripts/tools/build_multi_run_fixture.py script. 776K / 194 files;
# tracked by a conftest auto-generation hook rather than committed.
tests/fixtures/multi-run-corpus/

# lean-ctx local cache (per-user, not shared)
.lean-ctx/

# Virtual environment
# No trailing slash: the Makefile resolves $(PYTHON) from .venv/bin/python, so on a box that
# can only build the dev env (e.g. macOS x86_64, where .[ml] has no torch wheels) .venv is a
# SYMLINK to .venv-dev. "``.venv/``" matches a directory only, so the symlink showed up as an
# untracked file.
.venv
# Secondary env: ``make venv-dev-init`` (``.[dev]`` only, CI ``test-unit`` parity)
.venv-dev/

# Test outputs
.test_outputs/
# Real-episode eval datasets — real episodes are NEVER committed (only synthetic /
# fixture datasets like curated_5feeds_* stay tracked). Real-corpus manifests, silver
# references, and WER/bake-off datasets live here untracked. See feedback: never-commit-real-episodes.
data/eval/datasets/prod_*
# Real-episode-derived labelled sample (real insight text + corpus paths)
docs/wip/NER_FP_SAMPLE_LABELLED_2026-06-24.json
# Throwaway UI verification screenshots (chrome-devtools captures)
.wip-screenshots/
# Local prod backup verification (scripts/ops/verify_prod_backup_snapshot.sh)
.tmp_backup_verify/

# Stack test (Playwright under tests/stack-test/, ``make stack-test-*`` / CI): local logs, export, deps
.stack-test/
# Host copy of the ``corpus_data`` volume (``make stack-test-export``)
.stack-test-corpus/
.stack-test-corpus/
# Deprecated: full-run transcripts belong under ``.stack-test/`` (keep pattern so stray root files stay untracked).
.stack-test*.log
tests/stack-test/node_modules/
tests/stack-test/test-results/
tests/stack-test/playwright-report/

# RFC-064 profile capture runs (metrics + artifacts; frozen YAML goes under data/profiles/)
.tmp/
# CI metrics bundles downloaded for local validation (see scripts/dashboard/fetch_ci_metrics_artifacts.sh)
artifacts/ci-metrics-runs/
artifacts/nightly-metrics-runs/
# Unified dashboard data at repo root `metrics/` (generated in CI or fetched locally).
# Canonical public copies live on GitHub Pages; committing snapshots here causes noise and drift.
# Populate with `make fetch-ci-metrics`, `make fetch-nightly-metrics`, or workflow outputs — see docs/ci/METRICS.md.
metrics/*.json
metrics/*.jsonl
metrics/index.html

# Ad-hoc experiment batch log (machine-local; regenerate from your runs if needed)
docs/wip/_batch_experiment_log.txt

# Agent journal — persistent, machine-local working notes (plans/approaches/logs/
# decisions) that survive a session clear. Ignore all entries; keep only the README
# that documents the convention. See AGENTS.md → "Document location".
.journal/*
!.journal/README.md

# Local unified dashboard preview (CI bundle + nightly from metrics/)
artifacts/dashboard-preview/
output/
run_*/
# Streamlit run comparison app (path matches run_* above)
!tools/run_compare/
!tools/run_compare/**
# Committed consumer-app validation corpus (path uses the real run-dir layout,
# feeds/<show>/run_*/..., which the broad run_*/ rule above would otherwise ignore).
# This is a checked-in, deterministic fixture — see
# tests/fixtures/app-validation-corpus/README.md.
!tests/fixtures/app-validation-corpus/**
reports/
# Exclude dependency graphs (generated files)
reports/deps*.svg
reports/deps*.json
# Perf-trace per-release reports are tracked DOCS (eval-style history), not
# generated build reports — re-include them past the broad ``reports/`` rule.
!docs/guides/perf-traces/reports/
!docs/guides/perf-traces/reports/**

# Architecture diagram intermediates (only .svg outputs are committed)
docs/architecture/*.dot
docs/architecture/*.gv

# Benchmarks
benchmarks/

# Python cache
*.pyc
__pycache__/

# Crash dumps (false alarms from faulthandler during long-running operations)
crash_dump_*.log

# OS files
.DS_Store

# Coverage (pytest output at repo root only — the unscoped pattern was
# accidentally catching ``web/gi-kg-viewer/e2e/fixtures/production-shaped/
# corpus/coverage.json`` which is a Tier-2 mock-response fixture).
.coverage
.coverage.*
/coverage.json

# IDE
.vscode/
.vscode/settings.json

# GI/KG viewer (Vite / Node / Playwright, RFC-062) — was web/gi-kg-viewer/.gitignore; rules live here
web/gi-kg-viewer/node_modules/
# Catch-all: stray ``node_modules/`` at the repo root (or anywhere else) is always a
# transient artifact. The legitimate Node project lives under ``web/gi-kg-viewer/``; a
# root-level ``node_modules/`` typically appears when Vitest is invoked with CWD=repo
# root (e.g. ``npm exec --prefix web/gi-kg-viewer …``) — see VIEWER_TESTING_LOCAL.md.
# Always ``cd web/gi-kg-viewer`` before running ``vitest`` / ``playwright`` so caches
# land in the viewer's own ``node_modules/.vite/``.
node_modules/
web/gi-kg-viewer/dist/
web/gi-kg-viewer/dist-ssr/
web/gi-kg-viewer/logs/
web/gi-kg-viewer/**/*.log
web/gi-kg-viewer/npm-debug.log*
web/gi-kg-viewer/yarn-debug.log*
web/gi-kg-viewer/yarn-error.log*
web/gi-kg-viewer/pnpm-debug.log*
web/gi-kg-viewer/lerna-debug.log*
web/gi-kg-viewer/*.local
web/gi-kg-viewer/.idea/
web/gi-kg-viewer/*.suo
web/gi-kg-viewer/*.ntvs*
web/gi-kg-viewer/*.njsproj
web/gi-kg-viewer/*.sln
web/gi-kg-viewer/*.sw?
web/gi-kg-viewer/test-results/
web/gi-kg-viewer/playwright-report/
web/gi-kg-viewer/e2e-results.json
web/gi-kg-viewer/blob-report/
web/gi-kg-viewer/playwright/.cache/

# Playwright MCP (Cursor / IDE): local snapshots and logs, not repo artifacts
.playwright-mcp/

# AI rule-adherence log (local JSONL only; docs live in .metrics/)
.metrics/rule-adherence.jsonl
# Legacy path (if present locally)
.cursor/metrics/

# Cursor prompts (user-level templates live in ~/.cursor/prompts/)
.cursor/prompts/

# Example/test files (under config/)
config/examples/config.my.*
examples.txt

# Acceptance: track README + MAIN_ACCEPTANCE_CONFIG matrix + fragments; other *.yaml local
config/acceptance/*
!config/acceptance/README.md
!config/acceptance/MAIN_ACCEPTANCE_CONFIG.yaml
!config/acceptance/fragments/
!config/acceptance/fragments/*.yaml

# Local-only operator / feeds experiments (YAML etc.); tracked: dashboard import pointer
config/manual/*
!config/manual/README.md
!config/manual/grafana-dashboard-pipeline-execution.json

# CI-related local files under config/ci (optional stem lists, etc.); not committed
config/ci/*
!config/ci/README.md
# Stack-test seed lives at ``config/ci/stack-test-seed/`` and is consumed
# by ``make stack-test-seed`` to populate the ``corpus_data`` volume
# before the Playwright full UI flow runs.
!config/ci/stack-test-seed/
!config/ci/stack-test-seed/**

# Experiments (local / ad-hoc configs)
config/playground/

# Evaluation dataset (contains personal transcripts and reference summaries)
data/eval-old/

# Evaluation results
data/results-old/

# Evaluation materialized outputs (ignore contents, but track the README
# template — dataset-materialize requires it on disk to write per-dataset
# README files; without it CI fails before any model load runs).
data/eval/materialized/*
!data/eval/materialized/MATERIALIZED_DATASET_README_TEMPLATE.md

# Evaluation runs (ignore contents, but the marker note IS tracked).
# The marker warns future-you about the pre-fix scoreboards captured during
# the 2026-02-06 → 2026-06-22 silent-test-defaults regression window
# (see docs/wip/POST_RFC097_DEV_PROD_REMOVAL.md).
data/eval/runs/*
!data/eval/runs/_PRE_FIX_NOTE.md
# Shipped parity reports referenced by regression tests must be tracked, or the
# tests fail with "shipped parity report missing" in CI. Add new exceptions
# here whenever a `tests/e2e/test_*_parity_regression.py` grows a new fixture.
!data/eval/runs/v5_parity_2026-07-05.json

# Generated eval configs (scripts/eval/pipeline_validate.py auto-writes here)
data/eval/configs/_pipeline_validate/

# Generated search artifacts — regenerable, never commit the binaries.
# Two-tier LanceDB index (search/migration.py, #858) + trained query-router model
# (scripts/train_query_router.py, #860).
data/lance_index/
data/query_router.joblib

# Environment variables. Prefix globs (not exact names) so backup/variant droppings like
# `.env.obs.dev.pre-obs-sync.bak` are ALSO ignored — the exact-name patterns missed those and a
# real secrets backup got committed (2026-08-10).
.env
.env.local*
.env.autoresearch*
.env.*.local
# Dev observability secrets (GlitchTip DSN + Langfuse keys pointing at the homelab).
# Template committed as .env.obs.dev.example; the real keys stay local-only.
.env.obs.dev*
# Homelab obs-MCP deploy read tokens (template committed as .env.homelab.example).
.env.homelab*
# But NEVER ignore the committed *.example templates — they are legitimate and tracked.
!.env*.example
# Editor/backup droppings — never commit (they hid the leaked env backup above).
*.bak
*.orig
*.save

# Local ML model cache - ignore all contents except README.md
.cache/*
!.cache/README.md

# Local spaCy model wheels (see make download-spacy-wheels, docs/guides/DEPENDENCIES_GUIDE.md)
wheels/spacy/*.whl

# Autoresearch merged experiment configs (temp files next to score.py)
autoresearch/**/eval/autoresearch_cfg_*.yaml

# Autoresearch run logs
autoresearch/logs/
logs/

# Autoresearch local sweep ledgers — laptop iteration output; the weekly
# ledger (autoresearch-YYYY-WNN.json) stays tracked, but per-run local
# artifacts are transient (see ``make autoresearch-sweep-local``).
data/autoresearch_baselines/autoresearch-local-*.json

# pyinfra deploy debug log (transient per `make dgx-deploy` run)
infra/dgx/converge/pyinfra-debug.log

# Claude Code local settings (per-clone permission overrides)
.claude/settings.json
# Harness scheduled-task state — per-session, not project state.
.claude/scheduled_tasks.lock

# Worktree-specific files
WORKTREE.md

# Local dev-server runtime data (per-user files written while hand-testing the app)
.devdata/

# app_data_dir written at the repo root by ``make serve-for-validation``
# (SERVE_OUTPUT_DIR=repo root → <root>/.app): sessions, per-user files, audit +
# graph-analytics jsonl. Per-run runtime state, never committed.
/.app/

# Consumer Learning Player (Vue 3 / Vite / Playwright — RFC-099) — Node/UI artifacts.
# Mirrors the web/gi-kg-viewer/* patterns above. Was under app/* until slice 14.
web/learning-player/node_modules/
web/learning-player/dist/
web/learning-player/dist-ssr/
web/learning-player/logs/
web/learning-player/**/*.log
web/learning-player/*.local
web/learning-player/.idea/
web/learning-player/test-results/
web/learning-player/playwright-report/
web/learning-player/e2e-results.json
web/learning-player/blob-report/
web/learning-player/playwright/.cache/
web/learning-player/validation-results/
web/learning-player/e2e/.app-state/
# The viewer's equivalent. Since #1619 gave the viewer suite a real API, its webServer writes
# per-user state here exactly as the player's does — and unlike the player's, this path was not
# ignored, so a local run left runtime JSONL staged. It reached main in ef069c1e via `git add -A`.
web/gi-kg-viewer/e2e/.app-state/

# LCP trace raw JSON (large; keep only the .gz alongside)
docs/wip/graph-v3/traces/*.trace.json
# Search v3 eval — transient `make eval-search` output; keep only labelled
# baselines (S0-baseline.json, S<N>-<label>.json) checked in.
docs/wip/search-v3/eval/latest.json

.shots/

# MCP pivot-chain e2e: the app-validation fixture's two-tier search index is BUILT at
# test setup (offline, cached MiniLM) — not committed, to avoid a binary lance blob +
# lance-format-version coupling. See tests/integration/test_mcp_pivot_chain_e2e.py.
tests/fixtures/app-validation-corpus/v3/search/lance_index/
# Sidecar of the same index build (incremental-skip fingerprints). The rule above covered only the
# lance_index/ dir, so this artifact showed up as untracked the first time the index was built
# locally — see docs/wip/2026-08-13-e2e-on-intel-mac.md.
tests/fixtures/app-validation-corpus/v3/search/episode_fingerprints.json
tests/fixtures/app-validation-corpus/v3/search/metadata.json

# Upgrade migration lock (runtime, never committed)
tests/fixtures/**/upgrade_ledger.json.lock
scripts/eval/rolling_cache_*.json

# Disposable corpus copy the viewer e2e stack serves from (#1619). The operator plane WRITES into
# whatever corpus it is given — `GET /api/operator-config` creates `viewer_operator.yaml`, the jobs
# API creates `.viewer/jobs.jsonl.lock` — so `e2e/run-local-stack.sh` copies the fixture here and
# mounts the copy, leaving the tracked corpus (force-included at line 82) pristine. Re-seeded on
# every run; safe to delete.
web/gi-kg-viewer/.e2e-corpus/

# pytest tmpdirs that leak into the repo: the archive e2e tests pass an ABSOLUTE
# pytest tmp_path as an rclone base_path, and RcloneStorageBackend strips the leading
# slash (correct for a real remote, wrong for a local one), so cold-store objects land
# under the CWD. Ignored in both places because it also entered the docker build context.
private/