Cassandra Lab

Self-discovering leading-indicator engine — mining cross-vertical event chains for validated "A precedes B" rules.
Honest by design. Day-1 history is thin, so most chains will read "need more history" or "weak" until the event log deepens. We never present a low-confidence correlation as certain — every rule carries n, base rate, lift, and a Wilson 95% lower bound.
How it works
  1. Ingest time-stamped events across 6 verticals (USAspending, SEC EDGAR, NIH RePORTER, ClinicalTrials.gov, CPSC, FDA) into one unified temporal log.
  2. For each candidate chain A → B, test whether A reliably precedes B (same entity) within N days, at rate P, with lift over base rate.
  3. Backtest on the historical log; keep only rules clearing min-sample + Wilson-bound > base-rate.
  4. Emit live predictions for entities that just triggered an A-event.
Event log (last ingest 2026-06-21T09:50:38.285Z)
VerticalEvents indexed
usaspendingError: HTTP 525 for https://api.usaspending.gov/api/v2/search/spending_by_award/
sec98
nih195
trials106
cpsc179
fda134
Miner status
712 events 561 entities 9 hypotheses tested 0 validated 0 promising