Question 1

What is in a proof report?

Accepted Answer

A before-and-after for one instruction: the rubric score delta broken down by dimension, the evidence behind each finding, the eval case that demonstrates the behaviour change, and a version hash so the exact artifact is identifiable. A document a stakeholder can read, not a claim.

Question 2

Is this a certification?

Accepted Answer

No. A proof report is reproducible evidence, not authority. It shows what changed and lets anyone re-run it to confirm. We deliberately do not call it "certified" — the value is that the result holds up to scrutiny, not that someone vouched for it.

Question 3

Can a client re-run it themselves?

Accepted Answer

Yes. The report references the artifact version and the eval, so the same check can be run again and produce the same result. That reproducibility is what makes it billable trust rather than a screenshot.

Question 4

How is this different from a screenshot of a score?

Accepted Answer

A screenshot is a claim frozen in time. A proof report references the version and the eval behind every number, so the result can be reproduced rather than taken on faith.

Question 5

What if a client disputes a finding?

Accepted Answer

Every finding is tied to a rubric dimension and a line of the artifact, and the eval is re-runnable. The disagreement becomes something checkable rather than a matter of opinion.

Question 6

Are the figures on this page from a real client?

Accepted Answer

No. The sample report shown here uses curated example specimens, labelled as examples. Your reports are generated from your own audits.

Show clients your AI work got measurably better.

Reproducible evidence, not a claim.

A document, not a claim.

Score delta by dimension

Evidence behind every mark

A version hash

Re-runnable, not certified

Answers before you start.

Know which instructions are ready to run.

Follow the review loop as it ships.