Generative MetricsAudit · Diagnose · Verify

Why Generative Metrics

Automated AI consumers are not one system, and website compatibility is not one score.

Audit

Record what each named protocol observed.

Diagnose

Trace each production finding to evidence.

Verify

Re-run the protocol required by the remediation.

The problem worth solving

A page can look correct in a browser while a direct client receives an access denial, an empty application shell, an unresolved redirect, malformed structured data, or a policy rule that applies differently to different actors.

Those are concrete compatibility conditions. They can be observed, repaired, and re-tested — which makes them auditable. Accountability starts with evidence a CFO can read and a client can inspect, not an abstract readiness grade.

The product boundary

What the audit can establish

Generative Metrics audits acquisition, actor policy, representation, extraction, supported structured data, and strict date evidence. It links production findings to those observations and verifies remediations on later scans.

Every result ships a sealed evidence bundle: its SHA-256 digest is published with the report, and anyone can recompute that digest and check the bundle against it at the public verifier (/verify).

What it delivers instead is accountability: verification proof — whatever the outcome — for every remediation, and a white-label report agencies can deliver under their own brand.

What it cannot establish

It is not a trust oracle, a visibility promise, or a prediction that a provider will crawl, index, train on, rank, cite, recommend, or understand a page.

Design rules

Observation before judgment

Acquisition artifacts and protocol observations are recorded before findings or remediation are derived.

Actor and purpose stay explicit

Discovery, training, grounding, and user-action consumers are not treated as one generic AI bot.

Unknown stays unknown

Failed or insufficient evidence produces FAILED, ABSTAINED, INCOMPLETE, or INDETERMINATE—not a fabricated score.

A fix must be verifiable

Every remediation names the protocol and expected outcome a later scan must satisfy.

Why an integrated product still matters

Individual checks exist. The hard part is preserving the acquisition artifact, applying the correct actor profile, separating representations, carrying limitations forward, preventing experimental output from becoming production truth, and proving whether a named remediation actually cleared.

The product is the controlled evidence chain and the repeatable workflow—not ownership of HTTP, robots.txt, JSON-LD, or content extraction as isolated techniques.

What the product actually shows

Every interface label below is quoted verbatim from the shipped product — the words the interface uses are part of the evidence-first contract. Annotated captures of each of these surfaces need a signed-in session and are a follow-up; no screenshot is faked here in the meantime.

The result page

Every result reads the same way: “What this result means”, then “Do this next”, then “What this does NOT mean” — the verdict's meaning and its boundary, side by side, before any detail.

The per-actor count

The coverage counter prints the honest number: “N of M token-carrying requests reached the page at scan time” — and says so explicitly when any probe did not.

The bulk matrix

A bulk audit shows plain progress — “Progress: N of M completed” — with the per-URL states and no invented percentages.

The public verifier

The bundle checker states its own scope: “MATCH — the bundle digest is intact”, with the sentence beside it that a matching digest does not authenticate who produced the file.

Start with one target

Run a bounded audit, inspect the evidence and limitations, fix one production blocker, and verify that remediation under the same protocol set.

Create account Review protocols Review proof contract