Capture the exact claim
Write down the precise capability, workflow or performance promise before judging it.
Methodology
Evidence categories before conclusions.
VetAI Trust evaluates claims by separating what is asserted, what is supported by vendor evidence, what has independent support and what has been independently reproduced.
The evidence ladder
Not verified ≠ false.
A lack of independent verification is an evidence status, not an accusation. It means readers should know what is known and what still needs testing.
- 1Vendor claim
- 2Vendor-supported evidence
- 3Independent evidence
- 4Independently reproduced evidence
Core flow
Claim to evidence to test to status.
VetAI Trust tries to keep the evaluative chain legible. Different policy pages explain the rules around each step; the methodology page explains the assessment logic itself.
Check what supports it
Review papers, validation reports, primary sources and disclosed limitations.
Measure when review is not enough
Benchmark only when an independent test answers a real public-interest question.
Publish a scoped status
Readers should see what is supported, what is unresolved and what changed over time.
Evaluation workflow
How a review moves from claim to finding.
The workflow is intentionally inspectable. Readers should be able to see how a conclusion was reached and where uncertainty remains.
Capture the claim
Identify the exact wording, context, product scope and practical implication.
Locate supporting evidence
Collect public materials, papers, validation studies and any vendor-disclosed evidence.
Assess fit
Ask whether the evidence actually supports the claim, not merely a neighboring or narrower claim.
Compare baselines
Use relevant workflow, expert, specialized veterinary AI and frontier-model baselines where appropriate.
Test if needed
Design independent testing only when review alone cannot answer the public-interest question.
Disclose limits
Publish findings with methodology, caveats, conflicts and unanswered questions.
Concrete over vague
Claims must be specific enough to evaluate. Broad phrases are translated into testable questions before review.
Primary sources preferred
Press language, abstracts and marketing summaries are not substitutes for study design and underlying methods.
Baselines matter
A test result is only meaningful when readers know what it was compared against and why that baseline was chosen.
Policy links
Method is not the same as governance.
This page explains how evidence is assessed. It deliberately avoids duplicating the policy pages that govern independence, benchmark access, privacy and corrections.