Source read via streamed vault open + bounded staging copy (index-only duration/extension); encoded output walked from a bounded stream (new OggOpusParser.WalkAsync, byte-identical to the buffer oracle) and stored via streaming vault write. Adds parity tests.
AddEntryStreamingAsync now writes to a temp file in the same vault directory,
renames it into place (POSIX rename(2) — atomic on Linux), and updates the
index only after the rename succeeds. A client disconnect or I/O fault during
the write leaves the original backing file intact and the index unchanged; the
temp file is cleaned up best-effort on failure. Fixes the data-corruption
regression on the replace path where a cancelled write could truncate the live
backing file after the index update and FileMode.Create already ran.
Also filters OperationCanceledException from error-level logging in
RegisterResourceStreamingAsync — a normal client disconnect is not an error.
Two tests added to AudioStoreStreamingTests covering cancel and fault on the
replace path.
Processors now emit a ProcessedAudio plan with a streamed writer instead of a whole-file AudioBinary; vault writes stream via RegisterResourceStreamingAsync. Header parsing is bounded. Wave 2 (waveform/Opus) still re-reads the full file by design.
Player picks Opus when the browser can decode it and a sidecar exists (else lossless), injecting the sidecar before stream init; seek reuses the same format. Adds the Backfill-Opus bulk API endpoint + CMS action.
Background-job transcode (ffmpeg/libopus) after source store; pure C# Ogg
walker builds the 0.5s-bucketed granule→byte seek index + captures the
OpusHead/OpusTags setup header into a per-track sidecar in a new track-opus
vault. Best-effort, additive, regenerated on replace-audio.
On partial failure the old path deleted the original audio before
confirming the new write succeeded. Now: load old extension, register
new audio first (original untouched on failure), then clean up stale
backing file only on success and only when extension changed.
Swap a track's audio by EntryKey (metadata/release/position preserved, waveform regenerated); hide per-track remove on a release's sole persisted track so it can only be replaced or release-deleted.
Move the landed Phase 12 section from PLAN.md to COMPLETED.md; update DeepDrftAPI/Content/Public.Client CLAUDE.md for the WaveformVisualizer rename, per-track high-res datum + track-waveforms vault, track-cardinal fetch, popover controls, Ambient slot, and NowPlaying host.
Per-track high-res datum keyed by EntryKey in the renamed track-waveforms vault; computed at upload for all tracks, regenerable per-track via CMS, with a re-runnable backfill. Mix read path repointed so it keeps working.
Smooth the loudness contour (~50 ms envelope at preprocessing + decode-time, plus
smootherstep render reconstruction); retune wax↔waveform collision to bouncy/sub-unity
(no explosion/stuck/jitter); split the bubbles knob into fluid-amount + fluid-viscosity
(cohesion via uniform-only smin/wobble); retune scroll/gravity/heat/width ranges; make
the colour rotation visible and boost OKLab chroma; the controls bar now holds its
layout and hides only its knobs via a Visible parameter.
Adds ReleaseRepository/ReleaseManager (IReleaseService) for paged medium-filtered
release reads and Session/Mix satellite writes, UnifiedReleaseService orchestrating
vault+SQL, and ReleaseController (5 endpoints). Refactors WaveformProfileService for
configurable bucketCount/vaultName (backward-compatible) and adds the mix-waveforms vault.
Promotes brittle error-string literals to named constants (MixHasNoTrackMessage,
MixTrackNoAudioMessage) on UnifiedReleaseService.
ILoudnessAlgorithm strategy (RmsLoudnessAlgorithm first impl), WaveformProfileService
stores quantized byte[] sidecar in new MediaFileVault (profiles vault), wired into
UnifiedTrackService.UploadAsync; failure is logged and swallowed. WaveformProfileDto
and WaveformProfileOptions in shared projects.
daniel-c-harvey