Research registry
Validated runs, published with their limits.
A Research Run is a reviewed comparison between a controlled official reference endpoint and a candidate endpoint using the same immutable benchmark. Research evidence is experimental and never becomes a model-identity verdict.
Current status · RESEARCH PREVIEW
No validated Research Runs are published yet.
The comparison engine exists, but no real official reference/candidate production run has completed the publication gate. ModelTrust will not publish a mock result as evidence.
Publication criteria
- Same versioned probe set for reference and candidate
- Compatible secret-free artifacts
- Observed and experimental evidence kept separate
- Limitations and sampling window disclosed
- Bilingual report reviewed before publication
Every future research article will disclose
- Provider or anonymized provider
- Claimed and reference models
- Benchmark and date
- Observed evidence
- Experimental evidence
- Limitations
- Methodology link