[ahrazzle]

Closes #58.

VERIFIED — descriptive registry metadata: a fresh ClinicalTrials.gov API v2 query returned 592 unique records, retrieved 2026-11-05 America/Chicago (2026-11-06 UTC). Under the prespecified dictionaries, 155 records mention a selected biomarker bucket, 120 mention a platinum-resistance/refractory bucket, and 50 mention both. The report includes mutually exclusive phase combinations and all keyword/source-overlap/missingness counts. These are text mentions, including exclusions and background, not confirmed eligibility or trial results.

The query, fields and literal keyword rules were committed locally before acquisition. The Python standard-library script follows pagination, reconciles IDs/counts, checks source-version stability, and emits aggregate JSON only. The result folder contains the frozen certificate, byte-integrity manifest, scoped verifier and evidence statement. The dated report records the exact executed request, ClinicalTrials.gov processing timestamp, UTC/local acquisition interval, terms/attribution, transformations, commands and replay limits. Claims/sources/credits and continuation links are updated.

Validation: all 350 Python tests pass, including 40 new artificial-data audit tests; repository offline validators, integrity checks and existing educational checks pass. A second live query reproduced every aggregate count and source version. Independent GPT-6 Astra review found no material counting or interpretation issue. CI now verifies the frozen aggregate offline. The narrow .gitignore exception allows only the four reviewed result files; other result data remain ignored.

No patient records, trial-by-trial rows, copied eligibility narratives, private Asma data, investigator contact, owner-host access or Pilot 001 pipeline execution. Current registry updates can change later live counts; retained aggregates cannot independently reconstruct historical text classifications. The historical 590 figure was not reused or treated as today's expected count.

AI assistance: OpenAI Codex with GPT-6 Astra agents for method/terms review, implementation and synthetic tests, under human direction. Command-line publication credentials were unavailable, so the connected GitHub API published identical file trees; commit messages retain each original local commit ID/time and the report distinguishes publication metadata from prespecification. No merge requested.

[cursor]

Rebased this branch onto current main at af6cc9b (squash-merge of #61) so it should merge cleanly again.

After the rebase, the trial-landscape ledger entries were remapped to avoid colliding with #61: V28 → V29 (592-recruiting/interventional ClinicalTrials.gov snapshot) and K30 → K33 (ClinicalTrials.gov API/terms KNOWN claim). #61’s exploratory-plan claims stay V28 (Mann–Whitney design) and K30 (34-tumour frozen set) as on main.

CLAIMS.md counts are VERIFIED 28 · KNOWN 36 · CONJECTURE 23; CI pins 357 Python tests; index.html shows 87 labelled claims. Frozen snapshot numbers, certificate hashes, query URL, America/Chicago retrieval window, and aggregate-only rules were left unchanged during conflict resolution.