METHODOLOGY / VERSION 1.0
Evidence first. Duties second. No black box.
A bounded technical review of the transparency signals visible across a public website.
01 / Scope
A public-surface assessment.
Evidently inspects pages and media that an ordinary visitor can access. It does not authenticate, evade access controls, or infer private implementation details. The report is therefore a record of observable transparency, not a certification of the organization or its underlying AI systems.
02 / Evidence
Every conclusion starts with a source.
The scanner maps same-site pages, inventories referenced media, and captures relevant text around each asset. It checks available metadata and provenance declarations, including C2PA-compatible claims where present. Signals are deduplicated before scoring, while source URLs remain attached for review.
03 / Article 50
Duties depend on role and output.
Article 50 distinguishes providers from deployers and treats some content categories differently. We test machine-readable marking, viewer-facing disclosure, deepfake context, and public-interest text separately. Questionnaire context refines this map but never overwrites captured evidence.
04 / Scoring
Readiness, weighted by exposure.
The score begins at 100 and deducts points for supported findings. Weight reflects the apparent duty, the reach of affected content, evidence confidence, and whether a gap repeats. An A indicates strong observable controls; lower grades indicate progressively larger or more certain remediation needs.
Evidently is a technical readiness tool, not legal advice or a conformity assessment. Findings should be reviewed against your role, deployment context, and current regulatory guidance.
Run a public scan →