Model-agnostic prospective evaluation
Bring your model.
Freeze the claim.
Let biology decide.
GaiaLab's long-term role is not to insist that one model wins. It is to make biomedical model claims falsifiable under the same predeclared evidence cutoff, cohort, endpoint and chronology rules.
01
Submit
Model identity, version, bounded prediction rows, scores, evidence cutoff and exact hashes.
02
Seal
Prediction artifact, endpoint, baseline, abstention policy and future outcome boundary become immutable.
03
Wait
No target outcome enters rescoring, prompt updates, calibration or cohort selection before closure.
04
Read out
All wins, misses, abstentions and uncertainty are published under the frozen rule.
Non-negotiable rules
No target-outcome fields at submission.
No post-lock row insertion, deletion or rescoring.
No hidden failures. Negative and inconclusive results remain part of the record.
Models are compared only on the same frozen cohort and endpoint.
Timestamp proof establishes chronology, not biological correctness.
Submit a model → Submission schema → Prospective Registry →