Evidence may inform the record
Vendor-supplied documents, studies and clarifications may be reviewed and archived for editorial traceability. Submission does not guarantee publication or a favorable conclusion.
Data & benchmark policy
Benchmark integrity is part of the product.
VetAI Trust should be transparent about what vendors may see, what must stay protected to avoid gaming, and how public benchmark results may and may not be quoted.
Some rules on hidden benchmark cases and participant data apply to future benchmarks as well as any current pilot testing. This page avoids promising infrastructure or access controls that are not already visible in the repository.
Results need context
Benchmark outputs are meaningful only with named scope, metrics, versioning, date and method context. Raw numbers without that framing are easy to misuse.
Integrity can require secrecy
Where VetAI Trust uses hidden cases, answer keys or internal reviewer notes, those materials are protected to prevent benchmark reconstruction or gaming.
Reasonable access
- Status updates on their own submitted material where applicable.
- Methodology information at a summary level.
- Their own benchmark results where VetAI Trust actually runs such a benchmark.
- Notice of factual-error review and an opportunity to provide additional evidence.
Integrity-protected material
- Hidden benchmark cases or answer keys.
- Confidential submissions from competitors.
- Private reviewer notes or reconstruction-enabling per-case outputs.
- Any material that would make it easier to game a benchmark rather than genuinely improve performance.
Result usage rules
Benchmarks are not free-floating marketing badges.
Context requiredPublic references to a VetAI Trust result should name the benchmark or test, version, relevant date, tested scope and the metric used.
No scope inflationA narrow result cannot be stretched into a general claim about all tasks, all species, all settings or overall product accuracy.
No false endorsementA VetAI Trust benchmark result is not automatically a certification, endorsement, recommendation or proof of performance outside the tested scope.
No deceptive editingSelective presentation, cropped caveats or misleading cross-version comparisons are not acceptable uses of VetAI Trust results.
What VetAI Trust protects
- Test secrecy where hidden cases are necessary.
- Versioned methodology and visible change notes when the evaluation basis changes.
- Separation between vendor participation and editorial conclusions.
- Review paths for factual errors without giving vendors editorial veto power.
What does not yet exist in code
- No participant dashboard or vendor portal is present in the tracked site.
- No repository-visible benchmark contract workflow is implemented on this branch.
- Current public participation is email-first rather than portal-first.
Results can be revisited
- New peer-reviewed work can change an evidence status.
- Independent validation or a methodological flaw can change how a benchmark is framed.
- Vendor-submitted evidence can trigger reassessment, but not guaranteed reversal.
Privacy-aware research graph
- The present Research Radar data model is primarily vendor, product, claim and evidence oriented.
- Paper author names appear as bibliographic metadata rather than as the core entity graph.
- Future monitoring should avoid unnecessary person-centric dossiers about individual employees or veterinarians.
Related trust pages
Method, governance and corrections still matter.
This page covers benchmark material and quoting rules. Governance covers independence. Methodology covers evaluation logic. Corrections covers challenge and update flow.