What an evaluation report actually contains
Every evaluation produces the same document for every product, which is what makes two reports comparable. This page walks through it section by section, so you know what will be published about your product before you submit it. Reports are public - you can open a finished one from any leaderboard.
Download a sample report
An evaluation produces two documents. The public report is what buyers read and what the leaderboard links to. The builder report goes to the product owner and carries the factor-level detail behind every score, including the gaps.
Public evaluation report
What is published for a product: the overall score, the six category scores and the evidence a buyer needs.
Open the PDFFor buildersBuilder evaluation report
What goes back to the product owner: the same scores plus the factor-level detail and the gaps behind them.
Open the PDFRead a real one first
Every published evaluation is public and lives at its own URL - there is no gated sample, because a sample you cannot check is a brochure. Open any leaderboard, pick a product, and you are looking at exactly the document described below.
Browse published evaluationsAnatomy of a report
Header and run metadata
The product, the category it was evaluated under, the evaluation date, the product version, and the access tier we tested. All four are recorded because all four change - a score without them is not reproducible, and reproducibility is the point.
Overall score
One number on the 1.0-5.0 scale, derived from the category scores by the framework weighting rather than a flat average. The weighting itself is published, so you can see which dimensions carried the result.
Six category scores
Quality, Use Cases & Pricing, Security, Privacy, Sustainability & Reliability, and Impact, Ethics & Safety. These are the level most readers stop at, and they are where a product shows a shape rather than a number - strong on capability, thin on governance, or the reverse.
Subcategory breakdown
Each category resolves into the measurable dimensions underneath it - Access Control Mechanisms under Security, Inference Data Governance under Privacy. This is the level at which a finding becomes actionable for a product team.
Atomic factors and evidence
The bottom of the framework: a specific, verifiable data point, its score, and the source behind it with its tier attached - from primary testing we ran, down through third-party audits and your own documentation, to anecdotal community reports. The tier stays visible rather than being averaged away.
Cohort comparison
Where the product sits against every other product evaluated under the same framework. A 4.1 means little in isolation and a great deal against a cohort median of 3.2.
Limitations
What was not tested and why - a tier we could not access, a capability outside the framework, evidence we looked for and did not find. An absent limitations section would mean we claimed complete coverage, which no evaluation has.
What you can and cannot change
Before a report publishes you see it in full. You can correct factual errors - a mis-stated pricing tier, a control we recorded as absent that is in fact documented - and you can point us at evidence we did not find. Both routinely move scores, because both change the evidence.
You cannot change a score by disagreeing with it, and you cannot suppress a section. If a finding stands after review and you still dispute it, your response is recorded alongside it rather than replacing it. An evaluation the evaluated party can edit is not an independent evaluation.
How scores are derived
The full scoring scale, the five source tiers and the factor weighting are set out on the methodology page, and the operational steps - classification, testing, review, publication - on the evaluation process page. Read those two if you want to argue with a score on the merits; both are published in full for that reason.
Submit your product
Evaluation intake takes the product, the access tier you want evaluated, and the documentation you want us to read. The report that comes out is the one described on this page.