[S1] pipeline re-run needed: activate review-article detection (PR #136)

[rhowardstone]

Context

PR #136 (merged) adds:

• JATS article-type extraction in tier0
• publication_type field in tier1 (LLM + JATS fallback)
• publication_type TEXT column in protocols table
• UI: review article banner + relabeling in protocol card

What's needed

A full pipeline re-run on the A100 to:

1. Re-run tier0_extract.py (extracts article_type from JATS for all PMCIDs)
2. Re-run tier1_extract.py (adds publication_type to all predictions)
3. Re-run build_kg.py (rebuilds DB with 24-column schema)
4. Commit the new data/kg/atlas.db to deploy-render

Known review articles in corpus

• PMC11235738 (Qi 2024, "Modeling, applications and challenges of inner ear organoid") — confirmed review by user
  • Currently shows: "Recipe card · 13 signaling factors · 100% grounded"
  • Should show: yellow warning banner + "Review-derived summary"

Gate criteria

Review flag only shows when publication_type == 'review' in the DB. Until re-run, all existing cards show unchanged behavior.

Linked to Issue #8 S1 sprint — grounding queue accuracy also benefits from removing review-article reagent noise.

[rhowardstone]

Codex Supervisor — status check

No agent has claimed this yet. This is S1 sprint work per issue #8. Pipeline Engineer (A100): this is yours.

Steps required:

1. git pull origin master (now at 1a3189b after navbar fix)
2. Re-run tier0_extract.py → tier1_extract.py → build_kg.py on the full 5,119-paper corpus
3. Branch: feat/publication-type-rerun, draft PR immediately on first push
4. Supervisor will merge when CI green

Gate: review article banner must appear for PMC11235738 (Qi 2024, confirmed review). Currently all cards show unchanged behavior.

S2 (KGX re-run) and backfill_pitfalls are also unstarted — claim one at a time per issue #49 protocol.