Measurement validity
Examine whether an automated evaluation or grader supports a specified claim. Review the rubric, reference judgments, calibration, disagreements, and limits.
For assurance practices
Bring one recurring AI evidence gap. We can scope a specialist workstream that helps your practice obtain, interpret, and challenge behavioral evidence about a system.
Prefer email? Write to [email protected].
Possible first scopes
Examine whether an automated evaluation or grader supports a specified claim. Review the rubric, reference judgments, calibration, disagreements, and limits.
Test a defined permission, delegation, or approval boundary in a safe setting, using permitted action and effect records.
Reconcile a reported conclusion with inspectable traces, configurations, and changes. Identify what the record supports and where additional evidence is needed.
The partner determines which evidence its engagement requires and whether Fide’s competence, method, and independence are adequate. Behavioral testing may need management-system evidence or specialist domain interpretation alongside it.
Proposed responsibilities
Fide is responsible for its agreed technical contribution. The engagement defines criteria, workpaper access, qualified review, recipients, and the report issuer before work begins.
Agree the evidence plan, version records, rerun materials, access restrictions, reviewer challenge, and correction process. Where evidence is unavailable, conclusions narrow accordingly. Unfavorable findings remain part of the work.
Before a customer commitment, confirm procurement, security, subcontracting permission, reviewer capacity, liability, and commercial terms. Any opinion or certification belongs to the authorized issuer under the applicable engagement criteria.
If Fide materially helped design or implement the subject matter, its assessment of that work is collaborative validation. An independent opinion requires an appropriately qualified assessor genuinely independent of that work and the relevant organizational conflicts. Disclosure or a different internal reviewer does not by itself establish independence.
Read the independence policy →