[vjeux]

User impact

A quick flick toward a toast's edge snaps back on any toast up to 120 px tall (every one-line toast); only a 48 px drag dismisses it.

Problem and solution fit

1. The flick needed 48 px. A release dismisses past max(40% of the height, 48 px), or as a flick (over 1.2 px/ms) that also travelled 48 px: up to 120 px tall, the flick never decides.
2. The flick was timed by its handlers, which run a frame or two late: in Chromium, input spanning 33 ms is handled over 55 ms, so 1.4 px/ms reads as 0.84.

Expected behavior and authority

• Toast.spec.md FR4 and spec:AST-094 FR36 (current): a release decides on projected travel (travel plus the release velocity times a projection) past the commit point, with no travel floor of its own for a flick.
• Toast.doc.mjs: toasts "swipe-dismiss toward their configured top or bottom edge"; spec:AST-003 FR16 (a toast swipe dismisses only the toast) is kept.
• iOS 26.2: a notification banner dismisses on a 15 pt flick.

Smallest restoration

• A flick needs no travel of its own (the swipe passed the slop to begin) and is timed by the input events' timeStamp.
• Unchanged: every threshold, the drag visuals, the exit, the timer pause; the speed stays the average since touch-down (feat(Toast): the swipe follows iOS notification banners under a finger #8121 brings release velocity).

Evidence

Mobile Safari, iOS 26.2 simulator; each path has a named test (ring drawn by the test page).

Path	Clip
Before (main): a 46 px flick (about 20 ms) on a 56 px toast snaps back (real time, then ¼ speed)
After: the same flick dismisses; the exit is Toast's existing one, unchanged here (over within a frame on this simulator)
After: a slow 36 px drag still snaps back (cancel)

Chromium phone emulation, CDP touch, 56 px toast: 46 px / 33 ms and 30 px / 23 ms flicks snapped back and now dismiss; slow 40 px and 60 px drags are unchanged.

Regression tests: pen and touch flicks on a one-line toast fail on main; a late-handled 30 ms flick fails with only the floor change.

Rebased onto main, which carries #8050, as one commit.

Scope

• One defect is restored.
• Separable contradictory or unsettled tagalongs were removed or split.
• No new public API, visual representation, default, or interaction model is hidden under the bug-fix label.
• Public text and artifacts contain no internal Meta context.

Testing

Local gates at 32035ec:

• pnpm test:ui (16 workers): 16,458 pass (654 files), with no unhandled error; the node project (16 workers): 14,560 pass, its only reds two cache.test.mjs cases (see fix(cli): a crashed run's release install is recognized by the Node it ran on, whatever its name #8520).
• Root pnpm build, pnpm typecheck, CI=true pnpm lint (0 errors) and pnpm check:knowledge: clean. A merge-tree with main (6700c24) is clean.
• Size (architecture:component-style-authoring INV6, the CI's measurement): Toast's import JS 63,022 → 63,036 B gzip against main (9ee7457); CSS unchanged (10,143 B).
• The new tests fail on main (3): the pen and touch flicks on a one-line toast and the flick timed by its input events.
• iOS simulator: the three clips above (the cancel clip was recorded with feat(Toast): the swipe follows iOS notification banners under a finger #8121's per-path clips).

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
astryx	Ready	Preview	Oct 30, 2026 11:15pm UTC

[astracat-bot]

AI review status for this pull request.

Review status	Updated
🟢 Reviewed (for maintainers only)	Oct 30, 2026, 11:30 PM UTC

[github-actions]

PR Analysis Report

Current run: The current CI run concluded skipped. Current analysis is unavailable.

The exact-head preview is not available yet.

---

_{Generated by PR Enrichment workflow | View current CI run}

[astracat-bot]

The change works as described: a quick flick now dismisses a one-line toast, slow drags behave as before, and the new tests fail on main and pass here. Approval waits on one decision, because no current record settles Toast's swipe thresholds (spec:AST-003 leaves them to the component): should a swipe toward the edge dismiss as a flick at any travel past the 8 px slop, timed by the input events' timestamps, instead of also needing 48 px of travel? Any engineering owner, the author included, records that decision by commenting /approve-spec 6c39f1425a58b4693bf944d170732c5271e00017 on this pull request, or can point to the record that already governs it.

[Automated review]

[cixzhang]

Posted by Cindy's agent on her behalf.

Parts of this PR re-implement behavior that already exists, or is being specified, as a shared part of core. Converging keeps one implementation per behavior:

• Commit or spring back on release. core/src/Toast/useToastGesture.ts:25 decides on its own whether a release finishes or springs back. A shared release rule is being specified (AST-057 FR5 covers row swipe today); avoid a component-specific threshold model.

This is about where the behavior lives, not about the PR's intent.

[vjeux]

Agreed: one implementation.

• Commit or spring back on release: the release speed moves to the shared velocity tracker and release filter (AST-100, spec(AST-100): one velocity tracker and one release filter for every gesture #8312), stacked on the PR that provides it; only the dismiss distance and the flick speed stay in Toast, as its own tuning on that velocity (AST-100 FR3).

[vjeux]

The release speed moved to the shared velocity tracker in #8121 (a0dfe79), which stacks on the foundation that provides it; this PR stays on main's estimator so it can land on its own.

[vjeux]

Current records now cover the threshold: Toast.spec.md FR4 and spec:AST-094 FR36 decide a release on projected travel past the commit point, with no travel floor of its own for a flick, so this removes the 48 px floor (at 32035ec); the rest of the shipped rule stays the gap FR4 lists.

[astracat-bot]

Approval recommended: a quick flick now dismisses a one-line toast, slow drags behave as before, and the flick's speed is read from the input events' own timestamps. This follows the Toast component record's release rule (FR4, through spec:AST-094 FR36, which sets no travel floor for a flick) and spec:AST-100 FR4, and the new tests fail on main and pass here. The other gaps FR4 lists, the shared tracker's release velocity and cancelling a release that moves back, stay for the follow-up the description names.

[Automated review]