LABS DATA CONTRACT
Experimental records can be public without being canonical.
This contract exposes Labs representations and PREDICTED relation candidates with their methods and boundaries. It does not elevate them into VERIFIED evidence or canonical SmellBook relations.
Separate by design
Experimental representation records and predicted candidates do not alter canonical smell profiles.
PREDICTED ≠ VERIFIED
PREDICTED and EXPERIMENTAL records are not VERIFIED evidence.
No secret embedding claim
The public Labs contract does not expose or claim a proprietary learned embedding/model.
No Personal SmellBook input
Labs does not use browser-local Personal SmellBook state to generate prediction candidates.
Machine-readable Labs export
Experimental representations, declared experiment registry and PREDICTED candidates with transparent signals.
Evaluation & reproducibility export
80 candidate explanations plus deterministic run and coverage metadata. No accuracy, probability or confidence claim.
Review & challenge export
80 read-only review packets plus 5 deterministic challenge sets. No decision or promotion state.
Review context & provenance export
80 read-only context packets linking candidate review to claim-specific VERIFIED context, registered sources and gap markers without verifying the relation.
Review lenses & matrix export
5 deterministic lenses and 16 descriptive matrix cells with no ranking or priority semantics.
Labs Change Audit export
279 deterministic fingerprint inputs with truthful historical comparison state: NOT_COMPARABLE.
Retained audit snapshots
11 immutable retained record set beginning on 2026-09-29, plus explicit reference-only history where fingerprints were not retained.
Retained release comparison
279 audit records compared between the first two retained snapshots using only explicitly retained inputs.
Experimental records are not chemical fingerprints, molecular descriptors or formulation instructions.