01
The problem
Both homepages talk about testable systems and visible limits. Method copy is not evidence. We needed a fixed protocol with named inputs, inspectable outputs and a clear account of what it could not prove.
02
Inputs
The run uses a fixed list of public pages on last-word.fr and last-word.ai, once at a desktop viewport and once at a mobile viewport. Each target has a URL, locale, page type and expected viewport. No client data enters the test.
03
System
A headless Chromium session visits every target. It records start and completion times, HTTP status, final URL, canonical, language links, headings, page errors, console messages and failed requests. French and English pages go through the same checks.
04
Inspectable output
The run writes one JSON record for every page and viewport, followed by a review matrix. A SHA-256 manifest identifies the internal report and data bundle after the run. The public note includes a redacted record and the source bundle hash, while local paths and uncleared screenshots stay private.
Redacted public example
{
"target": {
"locale": "fr",
"kind": "home",
"url": "https://last-word.fr/"
},
"viewport": {
"id": "desktop",
"width": 1440,
"height": 900
},
"startedAt": "2026-08-25T11:29:24.284Z",
"completedAt": "2026-08-25T11:29:27.923Z",
"status": 200,
"finalUrl": "https://last-word.fr/",
"navigationError": null,
"consoleMessages": [],
"pageErrors": [],
"failedRequests": []
}Internal source bundle hash verified on 25 August 2026
64d1319ebda7b3016c3ed33971f31f1eda3d92882e6237feb63d05285a9a28fa
The full bundle remains internal because it contains local working paths and QA material that does not belong on the public site. The hash identifies the source used for this excerpt; it does not make the archive publicly downloadable.
05
Limits
This run describes what the browser observed during collection. It does not prove long-term reliability, coverage outside the sample or an effect on conversion. We did not submit the forms. Legal compliance and a full accessibility review are separate pieces of work.
06
Human review
The protocol collects evidence. It does not approve its own work. A separate reviewer checks the findings, distinguishes observed defects from assumptions and missing data, and either releases the next step or stops it. Material changes require a fresh run.