Writing · 27 August 2026
Reviewed the way we review everything.
This practice sells one discipline: critical designs are challenged independently before they are signed, and the challenge is kept. The claims on this website are critical designs. So before you read them as marketing, you should know they were put through the same machinery — and what that machinery found.
01 · THE PANEL
Independent models, and proof before opinion.
The site has been reviewed in rounds by multiple AI models — different vendors, different toolchains, each in a fresh session with no shared context and no sight of the others’ findings. Before any finding counted, each reviewer had to prove it had actually read the live site: quote values that exist only in the current build, and confirm the absence of strings deleted from the previous one. Not because reviewers are assumed dishonest — because a confident review of a page never read is indistinguishable from a real one by tone. It is only distinguishable by arithmetic.
The gate earned its keep. One submission arrived with invented file sizes, invented quotations and a confident graded verdict — a complete review of a website that did not exist. It failed the arithmetic, was logged with no vote, and when an almost identical copy arrived later, it was struck again. A separate reviewer, served a stale cached copy of the site by its own tools, quoted the evidence and refused to review at all — which is the correct behaviour, and it is on the record as such.
02 · THE VERDICTS
Amber. Not green. On the record.
The verified reviews did not bless the site. They graded it amber, repeatedly, and they agreed on why: the site explains the method better than it proves it. They also found concrete faults — small text that failed accessibility contrast when actually measured rather than eyeballed, a navigation that disappeared on phones, a numbering sequence that broke mid-page, and a stale social-preview image still carrying claims that had been deleted from every page — caught only because someone finally looked at the pixels instead of searching the text.
Every accepted finding was fixed and redeployed — and then verified by reviewers who had never been shown the findings list, one of which recomputed the contrast mathematics from the live stylesheet and matched the fix to the decimal. Findings that did not survive checking were struck individually, with the reason recorded. Nothing was averaged away.
03 · WHY PUBLISH THIS
Because the alternative is asking for trust.
A practice that says “nothing is signed off by its own author” should not exempt its own shop window. Publishing an amber verdict is uncomfortable; it is also what a kept record looks like mid-remediation. The one finding still open — inspectable evidence from client work — is being closed the same way, in the open, under confidentiality rules that are not ours to waive.
And to be precise about what this is not: AI reviewers are tireless, fast and usefully merciless, but they carry no liability and, as one of them demonstrated, they can invent. That is exactly why the access gate is arithmetic rather than trust, and why every decision that came out of these reviews — including publishing this note — was signed by a person.
Start here
Start with the portfolio you already have.
The same discipline is available for the decisions in your portfolio. Bring the plan as it stands.
The first read ends in writing: which of the eleven gaps are live in your portfolio, and the first gate that would expose them.