LearnNewsExamplesConnect
Frontmatter
assigneesneo-gemini-maryam
createdAtNov 8, 2026, 9:20 AM
updatedAtNov 8, 2026, 10:46 PM
githubUrlhttps://github.com/neomjs/neo/issues/20593
authorneo-opus-grace
commentsCount2
parentIssuenull
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[]
blocking[]
closedAtNov 8, 2026, 10:46 PM

The film tour's throw sometimes carries no momentum on a hosted runner

neo-opus-grace
neo-opus-grace commented on Nov 8, 2026, 9:20 AM

Context

#20451 decides the fate of workstation/WorkstationFilmTourNL; #19919 was folded into it for that purpose. The spec's three arms run off-socket, so it is adapter-eligible. Its hosted record still keeps it out of the required engine tier, so PR #20584 declares it in EXCLUSIONS (buildScripts/util/e2eCiSelection.mjs), with this ticket as the owner.

The Problem

Hosted run 42461321072, PR #20584 at 70b1e0e, shard 1/14, where FilmTour runs alone:

  • Attempt 1. The viewer-driven arm ("Start film tour runs all scenes…") failed at WorkstationFilmTourNL.spec.mjs:255, "the momentum carried the rows at least as far again as the hand": observed.y was 474 against the required 2 × requested.y = 960. The probe line reads observed {x: 356, y: 474} for a throw of requested {x: 360, y: 480}. The rows moved only as far as the hand, so the release carried no momentum. In the same job, the autoGates arm's throw did carry it: observed {x: 732, y: 1133}.
  • Attempt 2, a rerun of the shard. All three arms passed, and the throws read 1133 and 1051.

So one throw in four carried nothing.

Observation versus inference:

  • Observed: the zero-momentum throw and its numbers.
  • Inferred, not measured — the cause:
    • GridDragScroll#onDragEnd (src/main/addon/GridDragScroll.mjs:156-162) reads the release velocity from the moves of the last 100 ms (a Date.now() cutoff), and starts no momentum when fewer than two moves fall inside that window.
    • The film's throw samples every 33 ms (moveDelay: 33, moveSteps: 24, curve: 'throw', apps/workstation/tour/fiveBeatFilm.mjs:485), so about three samples sit in the window.
    • A hosted stall of about 70 ms before the release would therefore leave fewer than two.

The Architectural Reality

  • The throw. GestureDriver#executeGridDragScrollStep (apps/workstation/tour/GestureDriver.mjs), whose quartic 'throw' curve puts the speed into the last samples.
  • The momentum. GridDragScroll#onDragEnd calls autoScroll, which applies friction of 0.95 per frame. The total travel is about 20× the release velocity whatever the frame rate, so only the release sampling depends on timing.
  • The witness. WorkstationFilmTourNL.spec.mjs:246-255 pins observed.y ≥ 2 × requested.y, as #19584 designed it ("because that is what the beat claims").
  • The scope limit. #19584's Out of Scope excludes changing the addon or its friction.

The Fix

  1. Measure the release. On a hosted run, log the moves that fall inside the 100 ms window at the release; the film-probe line can carry them. That confirms or refutes the stall.
  2. If the stall is confirmed, make the throw's release land within the window even under a stall, without changing the addon. For example, the driver releases in the same task as its last sample, or tightens its last samples.
  3. After green hosted repeats, FilmTour leaves EXCLUSIONS and rejoins RUN_PATHS.

Acceptance Criteria

  • A hosted receipt names the release window's sample count for a throw that carries no momentum. Otherwise, the stall hypothesis is refuted, with what does explain the throw.
  • WorkstationFilmTourNL passes on 3 consecutive hosted engine-tier runs. It moves from EXCLUSIONS back into RUN_PATHS, with its measured duration (177.8 s) in e2eEngineDurations.json.
  • GridDragScroll's velocity rule and friction stay unchanged, per #19584's scope, unless the measurement shows the rule is itself wrong for a real user.

Out of Scope

  • The film's other beats, and the hop 4 classification (#19919).
  • The off-socket adapter gaps (#20587).

Related

#20451 · PR #20584 · #19919 · #19584 · #15252

Ownership: self-assigned. It owns a CI exclusion entry.

Live latest-open + closed sweep: checked the latest 20 open issues and closed searches ("film tour momentum", "GridDragScroll velocity release") at 2026-11-08T08:19Z, and read the Out of Scope of the same-surface closed hit #19584. No equivalent or prior exclusion was found; the closed #18104, #18099 and #8916 cover the addon's own lifecycle. A2A in-flight sweep: the last 15 messages hold no claim on this surface. Memory Core sweep: no prior hosted-momentum finding. Own open assignments: #20587, #20451, #20328.

Origin Session ID: 03506c7d-003d-47bf-9c85-0c87a27e3213

Retrieval Hint: "film tour grid throw no momentum hosted runner GridDragScroll 100 ms release window"

tobiu referenced in commit 885a5d5 - "ci(e2e): the Brain-gated tier retires; its specs run off-socket or locally (#20451) (#20584) on Nov 8, 2026, 2:24 PM
tobiu referenced in commit 8ff14e8 - "fix(workstation): release film tour throw without post-move stall (#20593) (#20601) on Nov 8, 2026, 10:46 PM
tobiu closed this issue on Nov 8, 2026, 10:46 PM