No universal readiness score
BioAtlas does not collapse scientific validity, licensing, compute, reproducibility and translation into one number. These are different questions and often require different evidence.
Evidence should reveal uncertainty—not hide it behind a score.
BioAtlas separates what is documented, what remains unknown, and what cannot be compared across scientific tasks.
Documented versus unknown
A field is marked documented only when the record contains a traceable source or explicit access statement. Unknown means the atlas has not recorded it; it does not imply absence.
Task-specific benchmarks
Benchmark claims retain their dataset, task or metric, evidence type and source. Structure prediction, docking, sequence generation and cell simulation are not ranked on a shared scale.
Record-level verification
A review date is shown only when the source record contains an explicit verifiedAt timestamp. BioAtlas never substitutes the deployment date for editorial verification.
Transparent recommendations
Recommendations are ordered using visible domain, term, access and deployment matches. The interface shows the matched signals instead of presenting a confidence percentage.
Corrections are auditable
Correction submissions can enter a GitHub issue workflow with the model, field, rationale and primary source preserved for editorial review.
Evidence coverage fields
01Primary publication
02Code access
03API access
04License or access terms
05Structured benchmark evidence
06Independent validation
07Experimental validation
The documented-field count is a completeness indicator, not a scientific quality score. A model with more documented fields is not necessarily better than a model with fewer.