Monster registry expansion Phase 1: WARN signals and quality firewall

[MEHuggins]

Purpose

Expand the source registry safely while fixing the false-positive item problem discovered in artificial-wombs, humanoid-robots, and space-force.

This is a stacked draft PR targeting repair/feed-health-clean. It does not modify main or merge PR #10.

What this adds

• 47 reviewed primary-source entries across six lanes
• new warn-notices-us lane with 19 official state WARN systems
• new automotive lane with regulator, permit, OEM, and labor sources
• thicker artificial-wombs, humanoid-robots, space-force, and quantum-security lanes
• runtime expansion overlay, leaving canonical feeds.yaml unchanged
• WARN table/list adapter that emits employer-level items
• item-quality firewall that quarantines CAPTCHA, JavaScript-wall, homepage, topic/category, signup, and pagination pages
• adapter-aware feed-health probing
• five unit tests

Clean-file guarantee

Exactly seven files differ from the clean repair branch:

1. .github/workflows/update-area-feeds.yml
2. feed-expansion-monster-phase1.yaml
3. scripts/apply_feed_expansion.py
4. scripts/content_quality.py
5. scripts/feed_health_expanded.py
6. scripts/update_feeds_quality.py
7. tests/test_monster_expansion.py

No generated feeds, health snapshots, index churn, archive output, or Python caches are included.

Safety and rollback

• canonical feeds.yaml is copied and restored during every run
• exact duplicate sources are skipped
• a new lane must contain at least three independent source hosts
• rejected candidates are written to health/feed-item-quarantine.json
• closing this draft PR removes the entire expansion without affecting production

Validation completed

• Python syntax compilation passed locally
• five focused unit tests passed locally
• compare against repair/feed-health-clean: 8 commits ahead, 0 behind
• clean seven-file diff verified

Required before production

1. Run the workflow manually on this branch.
2. Inspect all 19 WARN source results and the quarantine report.
3. Confirm the generated WARN items retain employer, date, location, notice type, and affected-worker fields where available.
4. Review degraded sources and near-duplicate source entries.
5. Confirm generated outputs remain outside the review diff or are isolated on a disposable evidence branch.
6. Merge or deliberately restack PR Add feed health observability and repaired source overlays #10 first.
7. Obtain Michael's explicit approval before any merge to main.