Architecture SEO / TSEP — 15 October 2026 Three distinct reports keep scopes explicit: TS16 = two priority pages; TS17 = two essential hub-to-guide links; TS18 = 254 public reference pages. All C within their declared measurement scope; two orphan candidates remain documented. No global health, ranking, certification, 44-control or independent-review claim. Protocol 0.1.0-draft.31; source revision 640d7f18621e7b518c5a20cf144fed7a8107adb5. The exact selected control texts are retained in protocol-controls-fr/en.html under CC BY 4.0; availability of the GitHub commit is not asserted. Dates: plan 2026-10-15T03:36:37.603722+00:00; CMS 2026-10-15T03:38:41+00:00; crawl 2026-10-15T03:40:09.245Z to 2026-10-15T03:40:44.938Z. CMS and deployed routing metadata stability checked at the campaign boundaries, not continuous content stability. Raw private server configuration is not distributed. Public URL metadata, source provenance and hashes are retained. 256 broad candidates, 254 successful reference pages, 253 sitemap URLs, 258 HTTP requests including two linked legacy aliases and two nested Expertise mapping probes (HTTP500). 252 reference pages reachable in their language; orphan candidates: English GEO media kit and /ns vocabulary (only self anchors). Depth null means no path within scope, not zero. Architecture FR/EN: depth2 from own homepages, 16/12 distinct incoming source pages including language-switch links, 15/11 within main elements. These counts describe a new source graph; the frozen 188-page archive and its depths remain unchanged and are not a comparable before/after experiment. Boundaries: anonymous source HTML, clean public editorial routes, explicit deployed routes and CMS/library registries, two home roots, no invented edges from crawl/sitemap seed URLs. Excludes query states/pagination, public proof files and exports, assets, private/admin/API/actions, external hosts and non-editorial development scaffold. No JavaScript-rendered whole-site graph or Google indexing result. Legacy redirects resolved to final destination without adding an HTML-navigation step. Self links and nofollow excluded from reachability. Reproduce graph: python3 verify.py (Python standard library; reads public-html-captures.tar.gz without extracting). Validate each TSEP JSON with the pinned source: python3 tsep.py validate report-ts18.json ; python3 tsep.py gate report-ts18.json --summary --lang fr. Expected gate: GO_WITH_RESERVATIONS, custom single-control scope only; three separate reports avoid scope mixing. C means measurement/control requirements satisfied, not defects absent. Old report remains ../architecture-seo-review-2026-10-14/report.json (TS18 NT then). Local candidate only: no production publication. Other pages were not changed. Future structure changes require a new dated capture; old evidence remains a snapshot. Native TSEP HTML reports embed their evidence for offline use (about 11 MiB for TS16/17 and 20 MiB for TS18). report-fr/en.html are lightweight linked overviews; the guide does not load these evidence bytes.