[miguel-heygen]

What

lint and check no longer warn duplicate_audio_track for audio clips placed end to end on one track when the first clip's start plus duration only lands past the next start because of float rounding. For example, a clip at data-start="21.03" data-duration="0.1" followed by one at data-start="21.13" used to be reported as layered audio ("a.mp3 at 21.03-21.1s, a.mp3 at 21.13-26.1s"), because 21.03 + 0.1 is 21.130000000000003.

Why

The rule compared raw float ends: a.start < b.end && b.start < a.end. Any end that rounds a step past the next start counts as overlap. In the reported pattern (music cut at 21.03 s and laid back to back), 712 of the 3,000 two-decimal durations from 0.01 to 30 s produce such an end. The warning then sends authors and agents to "fix" a timeline that plays correctly, and trains them to ignore the real overlap warning.

Reproduced on 0.8.170: a project with 0-21.03, 21.03-21.13 and 21.13-26.13 on track 9 gets the warning.

Related work

None found (searched open PRs and issues for duplicate_audio_track and audio overlap).

How

• Core already owns "one instant" comparisons for clip times: instantTolerance, sameInstant and the span helpers in clipFacts.ts, used by the runtime and the timeline commands. Lint cannot import core (core depends on lint), so those helpers move unchanged to a new @hyperframes/parsers/time-spans subpath. Both packages already depend on parsers, and the runtime already imports parsers subpaths.
• clipFacts.ts re-exports them, so every existing import keeps working.
• The rule uses spansOverlap, so ends within one instant of the next start do not overlap, and a real overlap still warns.
• The subpath is registered in package-subpaths.json and synced to package.json with bun run sync:package-subpaths; tsup builds it.

Test plan

• Unit tests added/updated: project.test.ts "stays quiet for back-to-back clips whose start plus duration only rounds past the next start" fails on main (the warning is reported) and passes here, 3 runs in a row (61 passed). The existing "warns when audible videos overlap on one track" still passes.
• Manual testing performed: the CLI from this branch runs lint on the reproducing project with no duplicate_audio_track (3 warnings, down from 4). Moving the third clip to start at 21.10 (0.03 s of real overlap) still warns.
• Also run: core src/runtime and clipFacts (2121 passed), the runtime CI suite (test:hyperframe-runtime-ci), CLI src/timeline (113 passed), tsc for parsers, lint, core and cli, check:package-cycles and check:package-subpaths.
• Documentation updated (if applicable): not applicable.
• Comments follow CONTRIBUTING.md "Comments": the moved helpers keep their comments; the test says why its numbers matter.

[github-actions]

Edit accuracy: accurate 271 (base branch 271), smooth 174 of those

The gate passes.
Smoothness is reported in the artifact, not gated. A case fails only if it fails 2 of 3 runs.

Quarantined, measured but not gated (0)

Unstable (1)

• textedit-none-px-r0-nested-z100: tracking 0, pressJump -, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true / tracking 0, pressJump -, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true / tracking 0, pressJump -, drop 0, reload 0, render 0.03, renderKey -, undo true, teleport true

[somanshreddy]

Clean at 47c1a19b: no blocker.

• The helpers move into @hyperframes/parsers/time-spans unchanged, and clipFacts.ts re-exports them, so existing imports keep working. The subpath is registered in package-subpaths.json, in both package.json export maps and in the tsup entry.
• I brute-forced the rule outside the repo with the same spansOverlap. That covered about 2.5M back-to-back pairs at 2- and 3-decimal starts and durations, each parsed the way the attributes are, with starts up to 3600 s and durations up to 60 s:
  • The old a.start < b.end && b.start < a.end check gives 314,892 false duplicate_audio_track warnings.
  • The new check gives 0.
  • A real overlap of one unit (0.01 s or 0.001 s) still warns in every case: 0 misses.
• Zero-length clips behave as before. One inside another still overlaps; one at the other's start still doesn't.

— Somu