Public prospective predictions

PubMed baseline

Empty set; PubMed query uses mindate >= snapshot_date + 1, so every hit is post-freeze by construction.

PubMed match criteria

PubMed esearch with term: (drug[tiab]) AND (outcome[tiab]) AND (therapeutic OR treatment OR therapy OR clinical trial)[tiab].

Criteria are intentionally strict to minimize false positives. Drug synonym expansion (e.g. FTY720 / fingolimod / Gilenya) is not yet applied — see the GitHub repo for the open issue.

ClinicalTrials.gov baseline

Computed retrospectively from clinical_trials_cache (populated 2026-05-12 evening) using last_update_date <= snapshot_date. PubMed baseline frozen at 2026-05-12; CT.gov baseline reconstructed post-hoc.

ClinicalTrials.gov match criteria

Case-insensitive equality on drug + outcome between snapshot hypothesis and clinical_trials_cache row. Delta = cache NCTs with last_update_date > snapshot_date and nct_id not in baseline.

First delta check returns 0 new trials by design — the cache was populated the same evening as the snapshot freeze, so last_update_date > snapshot_date is empty until the next monthly cache refresh. Status will accumulate organically.

Status transitions

• pending — 0 new PMIDs and 0 new trials since freeze
• partial_support — 1-2 new PMIDs OR any new trial registered
• strong_support — 3+ new PMIDs OR phase-2+ trial registered

Integrity

The frozen snapshot is committed to git as data/prospective/snapshot_v2_2026_05_12.json and is append-only. Every monthly check writes data/prospective/checks_v2_2026_05_12_YYYY-MM-DD.json. Postgres is a derived cache and can be rebuilt at any time from these files. Full source: scripts/prospective.