docs: handoff — quality profiles, what's left, and the ground truth - #57
Merged
Conversation
Captures where the quality-profile work stands, what is deliberately switched off and why, and the ground-truth facts that were expensive to establish -- several of which contradict what the code previously assumed. The section worth reading twice is the failure pattern: four separate times in this work, code that passed tests and deployed cleanly did nothing at all (coverage measuring 1000 of 5612 tracks; the quality floor placed at a stage the pipeline never reached; a duration gate on two of four matching paths; a health row that 500'd the dashboard). Each was caught by querying production after deploying, never by the test suite. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HmUiLHPmKoz215WAWV5eHe
The worker's tests generate and probe REAL audio files -- that is deliberate, since the bugs they guard (AIFF bit depth in bits_per_sample, VBR bitrate needing max(stream, format)) are all metadata-reading bugs that a mocked probe would hide. The worker image installs ffmpeg; the GitHub runner does not. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HmUiLHPmKoz215WAWV5eHe
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Pausing the quality-profile work here. This adds
docs/HANDOFF.mdso a fresh agent can pick it up without re-deriving anything, and points the README at it.Where it stands
Live on v2.17.0, parity 94.93% (5,163/5,439).
Built and shipped: the Radarr-style profile (floor / cutoff / target) with a settings UI, tiered search that descends
hi-res → 24-bit → lossless → 320k, the rechecker that hunts above what a track already has, and the relocator that swaps the file and re-points Lexicon'sTrack.location.Deliberately off:
relocation_enabled=0. Finding an upgrade nothing can install is worse than not looking —_lexicon_find_or_importshort-circuits on the existinglexicon_track_id, so the better file would land on disk unreferenced while Lexicon keeps playing the worse copy. With the flag off, no hunts are queued at all.What's left
quality_tier; the rechecker can't evaluate what it can't see.wrong_version, 57no_tidal_match, 70other.Why the ground-truth section exists
Several facts in it contradict what the code previously assumed, and each cost real time:
location,bpmanddateAdded— direct SQLite writes only.location-only write preserves cues, grids and cloud links (verified, not assumed).lexicon_track_idas TEXT while Lexicon'sTrack.idis INTEGER — comparing them raw matches nothing and silently degrades to path matching.bits_per_sample; without the fallback all 974 AIFFs drop a tier.CREATE INDEXon a new column in the sameexecutescriptsilently skips every statement after it.The most useful part is the failure pattern: four separate times, code that passed tests and deployed cleanly did nothing at all. Each was caught by querying production after deploying — never by the test suite.
🤖 Generated with Claude Code
https://claude.ai/code/session_01HmUiLHPmKoz215WAWV5eHe