A complete audit · published unedited · faults included

The sample is the operator's own site, and it scores a C.

This sample is chrisbrock.io, run by the person who built the tool, published exactly as it came out. Nothing was fixed first and no finding was softened.

Score: 58.1%44.75 of 77 resolvable points
Grade: CMaterial discoverability friction
Coverage: 77%9 checks inconclusive, 23 points unscored
Pages read: 2537 requests, 41 seconds
Evidence: 37hashed records, each openable
Backlog: 27items, ranked impact against effort

Open the report Why a C is the point →

Corrected 7 September 2026

This report previously named chrisbrock.io/blog and chrisbrock.io/reading as pages with over 400 words and fewer than three headings, and its remediation prompt said to add headings. That was wrong. Both pages carry an H1 and twenty-odd H3s. Our HTML parser was discarding every heading whose text is a link, which is how blog indexes are usually built.

The audit was re-derived from the response bytes the original crawl captured on 1 September, not re-run against the site. Those bytes are published with it. The site did not change; two things in our code did, and both moved the score:

the parser fix above, which corrected the finding; and
a rule adopted on 5 September that stops a check claiming a site-wide pass from a sample that stopped at its own page cap. This crawl read 25 pages and stopped at a 25-page bound, so nine checks that previously read complete now read inconclusive.

The score moved from 65.5 % of 92 resolvable points to 58.1 % of 77, almost all from the second change. The page cap has since been raised, and a fuller audit of the same site scores 67.9. We publish the smaller number because this is the audit that was signed.

The parser kept a single capture slot, so a link inside a heading (<h3><a>Title</a></h3>) took the slot and the heading was dropped. Every site we had audited was read that way. The published report is now checked on every build against a replay of its own evidence.

Corrected again, 8 September 2026

The score moved from 45.25 to 44.75 of 77. One check, AC‑02, was reading the first 600 characters of a machine-readable guidance file and reporting conclusions about the whole thing. On this site that meant it missed a real off-site link past the 600-character mark; on ours it did something worse and invented one, by cutting a URL in half so that https://citeu no longer matched our own domain.

We found it when adding a standards-compliant security.txt to our own site lowered our own score. The check now reads the whole file.

Everything a paying customer receives

The report

Every check, its rating, the operator's reason, the limitations and the evidence ids behind each finding, plus the page table and evidence index the client copy leaves out.
report.html →
The JSON appendix

The same data, machine-readable: scoring with its denominator and inconclusive list, all 37 evidence records with URLs, observation times, response headers and content hashes, the backlog, and the crawl limits.
audit-result.json →
The operator's decisions

The seventeen assisted checks a person rated, each with its reason and limitations. This is the file the sign-off applies to.
decisions.json →
The remediation prompt

A paste-ready brief for whoever does the fixing, built only from findings that carry evidence. It proposes no fix for anything the audit could not establish.
fix-prompt.md →

What this sample shows

It scores a C overall. Two dimensions score F, citation readiness and measurement readiness, and three more score C.
Three checks are inconclusive. One, the rendered-page comparison, could not run in the audit environment; that is recorded as a tooling failure and not scored against the site.
Two of the three are inconclusive because the operator did not supply his own intake fields. The report says so instead of guessing.
No percentile is quoted. The report withholds one until thirty distinct sites have been scored on the same rubric.

Everything a paying customer receives

The report

The JSON appendix

The operator's decisions

The remediation prompt

What this sample shows