Strict direct topic cleanup

Classifier: strict-direct-v1

Keyword decisions

• Total raw keywords preserved: 15,458
• Accepted for production: 444
• Held for brand, model, or commercial review: 374
• Rejected as non-direct or explicit drift: 14,640

Only approved direct component-family terms qualify. DataForSEO intent remains stored as evidence but cannot override the deterministic text gate.

Existing published articles

• Published before cleanup: 294
• Retained with direct scope and complete source coverage: 18
• Quarantined model articles: 8
• Retracted non-direct articles: 234
• Temporarily withdrawn for uncovered source-pack terms: 34

The retained set has four valid internal links per article and no links to a withdrawn page.

Vector index

• Qdrant collection: msg_topics_strict_v1
• Accepted keyword vectors: 444
• Topic clusters: 255
• Borderline similarity pairs held for review: 979

Verification

• Selected keywords outside accept: 0
• Active jobs outside accept: 0
• Published articles with uncovered center-topic facts: 0
• Broken internal links: 0
• Published articles passing schema, editorial, and source-pack validation: 18/18
• Unit tests: 52 passed
• Jekyll production build: passed