[pullfrog]

Closes #29.

Both chord sources dropped the end of each detected segment and emitted onset-only events, so _eventDuration sustained every generated chord until the next onset and filled the rests in between. This carries each segment's own end through detection and honors it during generation.

Changes

• chordr/audio_chords.py — detect_chords_from_chroma now returns {"t", "name", "end"} per segment. end is the run's own end (the next frame's onset, or one frame past the last onset for a run reaching the end of the analysis), so internal and trailing silence in the input survives.
• chordr/screen.js
  • _normaliseHarmonyEvents carries an optional end through to generation.
  • _eventDuration honors an explicit end instead of always holding to the next onset, clamped to the next onset and to lengthSeconds.
  • _harmonizeMelody emits end per window (min(winEnd, lastEnd)), so a silent window is a rest and the final window stops where the melody ends.
  • Authored chart chords without an end keep the existing next-onset behavior.

Tests

• Python: internal silence is a gap (events[0].end at the silence, not the next onset) and trailing silence is preserved.
• JS: explicit end stops a chord before the next onset; not extended by lengthSeconds; capped by lengthSeconds and the next onset; the audio path honors detected run ends; harmonization does not sustain across silent windows and stops at melody end.

Mutation-checked each new test against the old behavior to confirm it fails without the fix.

node --test tests/*.test.js → 106 pass; python3 -m unittest discover -s tests -p 'test_*.py' → 20 pass.

Docs (README, CHANGELOG, CLAUDE.md) updated and plugin.json bumped to 0.6.1.

  ^{｜ View workflow run ｜ via Pullfrog ｜ Using big-pickle (free) ｜ 𝕏}

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (3 files)

• CLAUDE.md — Test count updated (106→107)
• chordr/screen.js — _eventDuration: when explicit end is present, honor measured duration exactly (bypass 100ms floor) while preserving clamping to next onset and lengthSeconds
• tests/generate_arrangement.test.js — Regression test for sub-100ms explicit end

Previous Review Summary (commit 8506ea4)

Current summary above is authoritative. Previous snapshots are kept for context only.

Previous review (commit 8506ea4)

Status: No Issues Found | Recommendation: Merge

Files Reviewed (7 files)

• chordr/audio_chords.py — Added end field to detected chord segments; correctly computes run end as next frame's onset or one frame past last onset
• chordr/screen.js — _normaliseHarmonyEvents carries end through; _eventDuration honors explicit end with clamping to next onset and lengthSeconds; _harmonizeMelody emits end per window capped at melody end
• tests/test_audio_chords.py — New tests for internal silence as rest and trailing silence preservation
• tests/generate_arrangement.test.js — New tests for explicit end behavior: stops before next onset, not extended by lengthSeconds, capped at lengthSeconds/next onset, audio path honors run ends
• tests/generate_accompaniment_from_lyrics.test.js — New tests for silent windows not sustained and accompaniment stopping at melody end
• README.md, CHANGELOG.md, CLAUDE.md, chordr/plugin.json — Documentation and version bump

---

_{Reviewed by free · Input: 36.5K · Output: 1.9K · Cached: 115.4K}

[cchacs]

Needs fixing: a 20 ms chord becomes 100 ms, extending into the rest. Reproduced from the actual function.

[pullfrog]

Fixed the reviewer's repro: _eventDuration was flooring every duration to 100 ms, so a detected/harmonized segment shorter than that (e.g. a 20 ms run) was stretched into the rest that follows it.

Change in chordr/screen.js: when an event carries an explicit end, the duration is now returned as end - start (still clamped to the next onset and to lengthSeconds), bypassing the 100 ms floor. The floor still applies to the derived-end path (next onset / trailing chord), where it originally belonged. Added a regression test asserting a 20 ms explicit end stays 20 ms.

Verified: node --test tests/*.test.js → 107 pass; python3 -m unittest discover -s tests -p 'test_*.py' → 20 pass. Pushed to the PR branch (732c06c).

Task list (6/6 completed)

• Checkout PR branch
• Fetch and read review comments
• Verify findings against code
• Implement fixes or reply
• Test, review diff, commit
• Push and reply to threads

  ^{｜ View workflow run ｜ via Pullfrog ｜ Using big-pickle (free) ｜ 𝕏}

The previous finding is fixed: explicit 20 ms chords retain their duration, with a regression test added. No new blocking findings. Verified locally: 107 JavaScript tests and 20 Python tests passed.