Scores should explain evidence—not hide it.
Genesis separates direct observations, heuristic indices, review hypotheses, and work not performed. The public diagnostic is a bounded HTML sample, not a certificate of whole-site health.
Up to eight pages and 36 link requests per website.
The public scan inspects returned HTML, page-response status, response time, document size, metadata, selected structural signals, and a bounded set of public links. It does not run a full browser, execute your checkout, read private source code, or retrieve private analytics. A discovered form is not a tested form.
Discovered URLs are the bounded URLs encountered in this examination, not the full website inventory. A completed request means the bounded examination finished, not that every page and feature passed.
An index, coverage, completeness, and pressure.
| Measure | Current calculation or meaning | Boundary |
|---|---|---|
| Sample Health | Rounded arithmetic mean of the available measured-category indices. The categories use disclosed HTML heuristics; inferred reach, engagement, traffic, and AI-readiness indices are excluded. | Not a calibrated quality, ranking, revenue, security, accessibility, or whole-site health measure. |
| Discovered URL Coverage | 100 × unique examined page URLs / unique URLs discovered in the bounded crawl. | Not a percentage of all pages that exist. Unknown routes and authenticated areas are outside this denominator. |
| Evidence Completeness | 100 × (passed + failed + review checks) / (all listed checks − not applicable checks). | Not statistical confidence. Inconclusive checks and listed capabilities not tested remain in the denominator. |
| Finding Pressure | Sum of finding-severity weights, capped at 100: critical 28, high 16, medium 8, opportunity 2. | Not implementation risk. Actual change risk needs source access, the proposed change, dependencies, and validation. |
The numeric API fields named coverageScore, confidenceScore, and riskScore remain available for compatibility. In the evidence-ledger format, the displayed meanings are the ones above. Earlier saved reports are identified as legacy and are not silently recalculated.
One outcome per check, with its target and limitation.
| Outcome | Meaning |
|---|---|
| Passed | The specific narrow check passed. An HTTP success does not mean a functional workflow passed. |
| Failed | The check observed a failure, such as HTTP 404/410, a server-error response, or an absent title. A transient response requires rechecking before a repair. |
| Review | Context is needed. Missing optional declarations or unusual heading counts do not automatically prove a defect. |
| Inconclusive | The result cannot establish success or failure. Examples include timeouts, access restrictions, rate limits, unverified redirects, and unknown schema applicability. |
| Not tested | The diagnostic did not execute that capability or access its required source. |
| Not applicable | The check does not apply to the observed surface, such as image-alt checking on a page without image elements. |
Each check is counted once. Failed checks, affected URLs, finding groups, and proposed repair groups are different units; adding them would inflate an issue total. HTTP 401, 403, and 429 are not automatically classified as broken links. Unverified redirects and timeouts are not counted as healthy links.
The link pass rate uses only conclusive link responses. When no conclusive responses exist, the link category is unavailable and excluded from Sample Health. Error or challenge HTML is not treated as the intended page by the evidence ledger.
Presence checks are not quality or policy verdicts.
A canonical declaration is not universally required. A noindex instruction may be correct for an account or utility page. A JSON-LD block establishes presence, not valid schema or eligibility for a rich result. An empty alt attribute can be intentional for a decorative image. Confirm purpose and suitability before changing these.
The legacy category indices still summarize coarse presence and threshold heuristics, including title and description length, headings, metadata, landmarks, and document delivery. Their values are not interchangeable with evidence-ledger failure counts. Consult individual evidence and context, not an index alone, before buying or applying a repair.
Response time and HTML size are not field Core Web Vitals. Public crawler directives and semantic markup do not prove actual AI-search citations. Reach, engagement, traffic, and AI-readiness indices are inferred; authentic audience metrics and visibility require authorized data sources and observed results.
Technical references: Google canonicalization guidance, Google noindex guidance, and HTTP status definitions.
Matched budgets do not prove equivalent websites.
Comparisons use the same page budget, but page types, actual coverage, hosting, bot restrictions, and response conditions may differ. Category gaps require results for both sites; an unavailable value must not be replaced with zero to manufacture an advantage.
Repair clusters are grouped hypotheses, not verified root causes. A diagnostic evidence export records observations and limitations; it is not a repair certificate. Verified repair outcomes require a reproducible defect, an authorized scoped change, before-and-after tests, regression checks, and approval. Those steps must be recorded before a repair can be represented as verified.