The site's first test suite, the first free third-party measurements, and a grade re-derived with eyes on every surface

What was wrong before this entry: every assessment on this site, in every version, named the same debit on the same row — this website ships no test and no CI — and the grade for the site never moved because the debit never did. The operator read the Systems tab and drew the correct inference: a system whose grade cannot change has not been re-assessed. The first test this repository runs is the one that keeps its own numbers honest. The suite is plain Node — no test framework was added — and it caught a real defect on its first run: nine record anchors pointed past the end of a turnkey document that had been rewritten two days earlier. The cure is a history-pinned anchor form, path at commit and line, which the test verifies against that commit. The suite now gates the deploy: the tests run before the build, the leak gate runs before the upload, and a perf budget refuses a build heavier than the last measured one plus fifteen per cent.

The second change is the first third-party measurement on this portfolio. The operator's constraint was zero cost, so two free public instruments were run against the live edges and their results are published with their reports: Mozilla's HTTP Observatory graded guardian-arm.com A+ and this site C+, because this site's content-security policy still allows inline scripts, which its static render ships; SSL Labs graded both edges B, capped by the edge still accepting TLS 1.0 and 1.1, which is a one-setting change. A measurement the site publishes only when it flatters is not a measurement, so the C+ and the B sit beside the A+ with the cure named. None of these is a human audit, and the copy says so.

The third change is the assessment itself, re-derived with eyes on every live surface at two widths and an instrument run on the artifact: contrast of every graded text node against its painted background, interactive target size, horizontal overflow, heading outline and document language. This site's home, record and brochure pages measured zero contrast failures across two hundred and two graded samples, zero of ninety-four targets under twenty-four pixels, no overflow, and one heading of the first level per page; guardian-arm.com measured zero of twenty-one and zero of nine; enumamodel.com carries two twenty-pixel targets and one text at 4.02 to 1 as named findings. On that evidence Design moved from B to B+ — accessibility measured on the primary surfaces, a design system in use — and stays below A- because keyboard focus order and complete multi-step flows were not audited. Guardian-ARM moved from B+ to A-: on the public edge with a working product, signed builds, a published honesty floor, a self-red-team run to a shutout and now an external A+, held from A by the absence of an in-vivo adversary and of any human third-party review. This site moved from B to B+. The whole holds at B+, because the caps that hold it are unchanged.