Repository navigation
Investigate phone graph displacement after rotation in the strict motion gate #208
Description
The strict phone motion gate failed on reviewed OAuth/LOD composition 34eb7db20b52f519f60cdecf650986d780384a2a in CI run40597251568. New-user acceptance had41 passing cases and one failed case (plus the harness execution-failure record).
graph settles after real drags, edits and rotation 390 @motion measured horizontal displacement185.395922px during its17-second sample; the bound is1px. The owned two-node fixture performs a real persisted drag, edits type, rotates390×844→844×390→390×844, waits for physics atRest, then samples every animation frame. Runtime and test evidence must establish whether this is actual idle motion or sampling a still-pending viewport/camera transition. Physics rest alone may not establish camera/layout readiness; this is a hypothesis, not a conclusion.
Acceptance:
- Retain and inspect the failing frame series, screenshot and Playwright trace; identify the operation that moves the graph.
- Reproduce with controlled scheduling in an owned fixture and retain failing-before evidence.
- Correct the responsible runtime readiness/ownership or test observation contract without arbitrary delays, wider1px bounds, shorter17-second sampling, retries or missing metrics.
- Preserve real drag server readback, edit, rotation and all three viewport cases; zero-retry final browser/CI acceptance is required.
Tracked under #193. Claimed by the current agent for investigation; independent review requested. This failure is not attributed to OAuth or the LOD title-only correction without evidence. The separate pre-correction LOD camera failure is #205 / #207.
Activity
mvalancy commentedon Oct 27, 2026
Implementation is now reviewable in #210 at
7cbf292f95869980a1784017b0786293c269f7e6. The trace shows one camera resize jump followed by stable frames, with unchanged source positions. The correction observes persisted/rendered edit completion and exposes camera readiness separately from physics rest. A controlled ResizeObserver negative proves the boundary; all three local viewports pass without widening the 17-second sample or one-pixel limits. Final composition validation, independent review and CI are pending. This issue remains open until verified post-merge CI.mvalancy commentedon Oct 28, 2026
The reviewed correction (#210) separates physics rest from camera/viewport readiness. Held ResizeObserver scheduling proves that sampling starts only after viewport convergence; all three real drag/edit/rotation cases retain the 17-second sample and 1px bound. The original 185px affine camera jump and subsequent stable frames remain retained.
Verified integrated
devhead:780a7599d1d7949d5e0fac56091f0b07bb0e6e0d, complete treecffa8b8a7903b9609603f1c42978511d900678f6, identical to independently reviewed and tested source525ca4e5aa5bd4a918ad85165d0f2a0e8caf59ac.Final integrated CI and Comprehensive Tests passed every applicable job. Downloaded strict reports contain smoke 6, new-user 42, LOD 2, routes 7, mobile 27 and performance 3 passing cases, with zero failures, warnings, skipped cases, evidence errors, timeouts or nonzero exits. Both supported Node versions, real Neo4j, all three runtime images, effects and showcase passed. The main/nightly-only production job was inapplicable to this dev run; it is not credited as a pass. Earlier failures and cancelled intermediate runs remain retained.
Integration used the owner's standing dev-only native admin authorization after independent exact-head review; that is not a non-author GitHub approval. Main remains
775362eace0723c57b7ba3c7936645c9d0d36cb4. No release, deployment or protection change occurred. Closing this bounded implementation issue after final branch acceptance, not claiming hosted or deployed-site clearance.