Wait for completed fragment navigation before scene qualification

Author: KSemenenko

docs/ADR/ADR-112-independent-website-publication.md

@@ -112,3 +112,5 @@ until actual measured and content-only qualification, freshness and Pages proof
 TASK-WEB-TRIGGER-SCOPE-001 implements REQ/AC-BC-WEB-008. The current owner direction supersedes this ADR's unfiltered push and success-or-failure final-dispatch clauses: `website.yml` filters trusted main pushes to `site/**`, its native site tests, site-specific scripts/composites and shared build inputs. `benchmarks.yml` retains the same final bounded dispatch with `actions:write` confined to that job, but joins all preparation/build jobs, every database matrix and aggregate; any failed, skipped or cancelled prerequisite prevents dispatch. Website's existing authenticated optional producer wait also requires the actual terminal conclusion `success` before qualification. No `workflow_run` subscription, alternate executor, measurement bypass or changed publication gate is introduced.
 
 Ordered root stages are policy/REQ/AC freeze, exact workflow edits, source/self-review and YAML/actionlint validation, then commit/push and inspect genuine trusted-main Website event/qualification/Pages evidence. The original bounded producer wait must admit a successful producer and reject a failed producer in actual Linux execution; this is required provider evidence, not a claim from static inspection. Push path matches and dependency outcome rejection are source checks; GitHub scheduling and successful producer dispatch remain explicit required provider evidence. Independent site changes can still publish without benchmark figures when no authenticated producer is ready. Original dated receipts remain historical. Rollback preserves the independent Website executor and all source/metric/coverage/browser/freshness/permission checks; it does not restore an unfiltered or failed-producer trigger.
+
+TASK-WEB-NAVIGATION-001 retains this architecture and tightens the existing AC-BC-WEB-003 browser operation oracle. The native menu action and restored fragment must finish their actual CSS-adjusted target scroll, clamped to the document's real scroll range, before subsequent hero navigation. The failing baseline is original Linux run37753944709/source284dd842: a real WebGL2 canvas was paused outside the viewport at `#product` after the mobile interaction. Test helpers own bounded observation of location and geometry; they cannot force renderer state, accept paused output, increase timeouts or lower coverage. Root joins the assertions/tokens and feature mapping, runs the canonical local site flow plus full quality checks, and retains the next original Linux qualification/Pages outcome. This changes only test sequencing and observable-result assertions; existing production, protocol, metric and provider boundaries suffice, with no new dependency or ADR decision.

docs/Features/BenchmarkComparisons.md

@@ -25,6 +25,8 @@ optional metric selection and independent Website publication.
 
 The owner's 2026-10-08 trigger correction is TASK-WEB-TRIGGER-SCOPE-001 under ADR-112: REQ-BC-WEB-008 requires automatic Website dispatch for actual site/qualification/build/publication changes and a successful Benchmarks producer. AC-BC-WEB-008 requires trusted-main site pushes to launch Website, unrelated backend-only pushes to omit it, and the final producer trigger to admit only successful preparation/build, all database matrices and aggregation. Failed, skipped or cancelled producer dependencies do not dispatch. A supplied producer run must retain exact own-repository/workflow/main/source admission and reach terminal `success` inside the original bounded wait. Manual Website dispatch and independent site publication remain available. Root owns `website.yml`, `benchmarks.yml`, policy/docs and YAML/actionlint checks; genuine GitHub event/job/Pages evidence and the actual bounded producer admission in Linux are the required manual provider exception for the trigger graph. No source-only check proves a GitHub dispatch or publication. All metric/archive/browser/coverage/freshness gates remain unchanged.
 
+TASK-WEB-NAVIGATION-001 strengthens AC-BC-WEB-003's existing real Chrome flow under ADR-112. The mobile menu action must close its popover, reach the actual `#product` fragment and finish scrolling to the target's computed CSS scroll padding/margin and clamped document scroll range before the next action. After leaving the document and returning to `#benchmarks`, the test must observe that fragment at its actual target and the lazy poster with no canvas before navigating to the hero. The existing visible-scene ready/unsupported, frame, poster, same-origin requests, empty content, coverage and joined cleanup assertions remain mandatory. Original Linux Website run37753944709 on284dd842 showed the restored renderer paused outside the viewport at `#product`; a closed menu and `document.readyState` alone did not prove completed navigation. Root owns the narrow test assertions/tokens, records this failing baseline, then runs the canonical site development flow, full build/format/governance and genuine Linux Website/Pages gates. No production renderer, timeout, allowed-state, dependency or coverage-threshold change is authorized by this repair. Three workers already own independent database movement scopes; root integrates this small shared browser-flow correction directly.
+
 | Requirement | Measurable acceptance | Tests and evidence |
 |---|---|---|
 | REQ-BC-CURRENT-001: the complete current native cohort is the sole live producer | AC-BC-CURRENT-001: the current plan accounts for all 1,386 workers across the 330-cell control, two 132-cell scaled profiles and 24 33-cell vector profiles, bound to the exact Benchmarks run/attempt/source and authenticated inputs. Each slot is an authenticated measurement, an explicit unsupported-topology disposition or a terminal workload failure with fixed safe reason, null report and visible original failure; measured workers pass correctness/fairness/provenance checks. Missing, malformed, expired, mixed, skipped, unaccounted or mismatched evidence rejects | SiteIsolatedGitHub selection/proof/archive/input/authority tests; actual aggregate and original GitHub archives |