#91 batch finished — all 60 famous tracks fetched + whole-mix analyzed.
First honest baseline on the full corpus (was n=5 before):

  top-1 emotion hit : 22%   (47% w/ overlap)
  V/A quadrant hit  : 42%
  mean V/A error    : 0.573

The center-compression hypothesis is confirmed at scale: 'playful' is the
predicted top emotion for ~18 of 60 tracks (Fleetwood, Marvin, Coltrane,
Avicii, Journey, Eminem...) because the distribution-weighted mean of the
circumplex anchors collapses toward the warm-positive centre.

Affine de-compression (gt ≈ a·pred + b, leave-one-out CV) learns exactly
that bias and corrects it:
  valence:  ×1.03  −0.25   (reads +0.25 too WARM — the playful pull)
  arousal:  ×1.70  −0.39   (compressed ~1.7x — needs the most stretch)
  LOO V/A err 0.573 → 0.519,  quad 42% → 50%.  Modest but real & cross-validated.

calibrate.py now writes per-mode files (calibration.whole-mix.json /
calibration.stem.json) so the #92 whole-mix-vs-stem decision can compare both
without one clobbering the other; calibration.json mirrors the freshest fit.

Stem-aware batch (demucs, all 60) launched in background to settle #92 on the
full corpus, not the n=5 that PLN rightly refused to conclude on.

New armada/tasks/completed-archive.md — per-task entries with description/done/learnings/
deps + commit hashes, richer than board-archive.md's terse snapshot, written as blog/video
source material. Seeded with this session's completed tasks: #65 (serve.py broken-pipe fix,
d70c3ee7 — keystone audition QOL) and #64 (Information-is-Beautiful dataviz epic + its
craft learnings). Appended by the /pre-compact ritual before clearing the board.

MidiStream fans one aseqdump subprocess out to N web subscribers (SSE), reusing
midimon.parse_line + enrich (one parsing source of truth); reader runs only
while watched (lazy start / stop on last unsubscribe, so no MIDI port held idle).
parse_ports is pure + tested. NOT yet wired into server/UI — see resume task.

Parked mid-feature: server SSE endpoint + dashboard panel still to do.

Clicking L'Armada served armada/ root → a file listing (felt broken). Point it
at armada/tide-table (where the viz lives) and open /triangle.html — same as
`tide.py serve`. Verified: /triangle.html → 200.

Also lands midimon.enrich() (note/velocity/controller→structured) + tests, the
parsing groundwork for the live MIDI dashboard panel.

The hub was dead links (clicking Foundry 404'd — its server wasn't running).
Now it's a real control center, with one shared registry driving both faces.

- launchers.py: allow-listed registry. Web tools (Foundry/Armada) = start-or-open
  via port-check (spawn the server if down, then open); native apps
  (Pulsar/Ardour/qjackctl) launch detached with running + availability state.
  Wayland has no portable focus → running apps report 'already-running'.
- midimon.py: 'aseqdump | tr' done right — parses each event, renders aligned &
  colourised with note names (C4) and velocity bars. parse_line is pure/tested.
- Bridge UI: hub shows live running dots; start-or-open opens a blank tab within
  the click gesture then navigates once the server binds (no popup block).
- perf-tray: 'Open Bridge' + 'Launch ▸' submenu off the SAME launchers.py — the
  tray is now the unified ParVagues tray (DRY with the web hub).
- 17 tests (launcher registry/state + MIDI parser). Verified: Foundry card boots
  the server → :8765 200.

So 'parvagues' is findable in the KDE app menu (the Bridge runs headless as a
systemd service → had no .desktop, nothing to index). Entry opens the dashboard;
ParVagues two-tone wave SVG as the icon. Installed via symlink into
~/.local/share/applications + kbuildsycoca.

Corpus 22→60 tracks, all 4 V/A quadrants now ≥13 (was +-:4, --:2 thin), 50 genres,
all 12 emotions — chosen for confident, balanced priors (quality of pseudo-GT > raw
count for calibration). Added the sad/dark-calm + calm/pleasant fillers (Portishead,
Mazzy Star, Satie, Debussy, Coltrane, Bon Iver…) the gaps report flagged.

emotion_calibrate.py: fits per-axis affine gt≈a·pred+b over analyzed∩GT to DE-COMPRESS
the center-biased CLAP read (learns the scale-up from data, not a guessed tau).
Reports leave-one-out CV error (honest generalization, not in-sample) + re-derives the
label from calibrated V/A (nearest anchor) so we see if labels improve too. Pragmatic
call: scale corpus = whole-mix analysis (demucs doesn't scale to 100s/1k; stems didn't
beat mix on accuracy) — stems reserved for the hexa lane.

- perf-tray.py drops its duplicated Thermals/detect_mode/_read*/MODES/SCRIPT
  and imports them from tools/bridge/perf.py — the tray is now just the Qt face
  of the same logic the web Bridge serves (one source of truth)
- tray icon restyled with the ParVagues signature wave (drawn in Qt, no asset)
  under the heat-coloured temp badge
- parvagues-bridge.service → WantedBy=default.target so it boots headless with
  linger (always-on dashboard, no login needed); tray stays graphical-session
- deployed: bridge enabled+active on :8773, tray restarted, linger enabled

PLN's validation idea: famous tracks whose feeling/structure/genre the model already
knows = pseudo-ground-truth to grade the emotion engine, disagreements driving tuning +
coverage (self-correcting 30→100→300). emotion_corpus_gt.json: wave-1 22 tracks spanning
all 12 emotions + V/A quadrants + genres + structural archetypes (steady / quiet-loud /
long-build); GT = emotion labels + V/A centre + approx fractional section structure +
genre. emotion_corpus.py: fetch (yt-dlp android+bestaudio 403 workaround) → analyze
(demucs+stem-aware or --mix) → grade (emotion-hit, V/A error, quadrant, vocal-presence) →
gaps (coverage histogram → next wave).

BASELINE (whole-mix, 5 tracks): 40% emotion-hit, 40% quadrant-hit, mean V/A err 0.75 —
honest verdict: NOT yet trustworthy. Three systematic biases surfaced: (1) center/positive
compression (extremes pulled to mild +V/+A), (2) 'playful' over-fires (3/5), (3) whole-mix
misses aggression entirely (Prodigy Firestarter→playful +V) — the exact case stem-aware
(drums/bass arousal) should fix. Corpus now drives calibration before the #90 hexa handoff.
gaps: +- and -- quadrants thin, fill at wave 30.

Adds analyze_stems(): separates via foundry_stems (vocals/bass/drums/other), scores
each stem on a shared window grid, fuses V/A by stem role (valence rides vocals+other,
arousal rides drums+bass+mix), and flags vocal presence per window as a structure cue.
Refactor: shared _finalize (smooth→relabel→section→summary) now serves both whole-mix
analyze() and analyze_stems(); embed_at/window_grid/windows helpers extracted.

FINDING (Bohemian Rhapsody, mix vs 4-stem): stem-aware gives a cleaner valence track +
vocal-presence sections (88% vocal; outro correctly reads calm, not 'cold'), BUT the
current arousal fusion UNDER-weights the mix and softened the hard-rock peak that the
whole-mix caught (tense A+0.77 → hypnotic A+0.55) — that section's intensity is a
full-mix gestalt isolated stems lose. So fusion weights need empirical tuning; the
ground-truth corpus (next) drives it. Both modes run end-to-end (#89).

Answers 'is there a ParVagues web service running a dashboard?' — now yes.
Folds perf-tray's function into a web toolbar + extensible fleet hub.

- perf.py: rootless Thermals (hwmon) + detect_mode + sudo-backed set_mode,
  ported from perf-tray.py with Qt dropped (DRY-able with the tray later)
- server.py: stdlib http.server (Foundry lineage); GET/POST /api/perf, /api/hub.
  Binds 127.0.0.1 by default — mode-switch flips the CPU governor via sudo, so
  not LAN-triggerable unless --host 0.0.0.0
- ui/index.html: Ship's Bridge toolbar (live temp badge 55/70/85, mode switch,
  freq/fan/throttle) + hub cards (Foundry :8765, Armada :8731)
- bridge.py CLI (serve/status/perf); parvagues-bridge.service (systemd --user)
- 6 tests on pure perf logic; README incl. deploy + security notes

Smoke-tested: tests green, API live, mode detect + thermals reading correctly.

Stem-aware emotion needs the voice isolated: the full mix is muddy for CLAP, but
the VOCALS carry valence, DRUMS+BASS carry arousal, and vocal presence is itself a
section/structure cue. ensure_stems(file, slug) consumes the Foundry's in-process
Python API (tools/foundry, the shared separation engine) → {mix,vocals,bass,drums,
other} absolute paths, idempotent via catch.json (re-runs free). Validated: 4-stem
demucs separation of a 6-min track (Bohemian Rhapsody) lands all four stems clean.
Feeds emotion_timeline's stem-aware analysis (#89) and the hexa lane (#90).

- README: 'Use as an API' — Python import / CLI subprocess / HTTP API, the
  path+catch.json contract, env overrides (for sibling tools e.g. the #89
  emotional-arc tool that consumes the demucs engine for stems)
- README: add 'link' to CLI reference
- research/{A,B,C}_*.md: the three strand reports (empirical profile,
  fundamentals rulebook, prior-art + recommended zero-dep stack)
- research/PHASE2_FINDINGS.md: strand-D synthesis — corrected per-stem model
  (loops cut per stem, often-but-not-always time-aligned), loopability rubric,
  Take/StemSlice model, correlation engine (§5·0 GT-at-scale), Hall of Fame
  tierlist, section-label grounding (↔ #89), 80/20 build order
- PHASE2_LOOPFINDER.md charter: grounded in the real flow + ground-truth pairs

Windows a master (~12s/50% hop), CLAP-embeds each window (new sample_semantics.
embed_audio_arrays seam), runs the emotion_ontology kernel per window → (valence,
arousal, top-emotion) series; median-smooths, relabels from the smoothed circumplex
point, fuses short runs into emotion SECTIONS. Emits frames+sections+summary JSON —
the continuous-V/A + discrete-scene lane hexa wants. Validated on Bohemian Rhapsody:
caught the warm-intro → energetic-middle → tense rock-peak (A+0.77) → calm-outro arc.

Each emotion is a CLUSTER of angle prompts (adjective pairs + scene metaphors +
bodily framings, EN+FR) marginalized to one robust per-emotion score (mean cosine,
size-invariant) — the synonym→emotion fold mirrors the resolver's descriptor→family.
12 emotions × 6 angles, each placed on Russell's valence/arousal circumplex, so the
distribution yields a continuous V/A point + a discrete top-emotion label.

Per-window kernel: score(ae_row) takes one CLAP audio embed → {scores, dist, top,
valence, arousal, confidence}. Validated: probe archetypes land correctly (melancholic
0.76, aggressive 0.94, hypnotic 0.94, cold 0.72); real-audio path confirmed (pad→
dreamy/melancholic calm). Feeds the planned per-track emotion timeline (hexa lane).

- engine/publish.py: link_kit() symlinks Samples/<kit> → Dirt-Samples/<kit>
  so a freshly-cut kit loads as s "<kit>" (325/508 Dirt-Samples entries are
  this pattern). Idempotent; repoints stale links; refuses to clobber real dirs.
- foundry.py link <kit> CLI; 5 new tests (15 total, all green)
- PHASE2_LOOPFINDER.md: research charter for the loop-finder (Sonnet-agent
  deep-research pass) — grounded in the REAL flow (yt → separated/htdemucs →
  hand-cut → Work/Sound/Samples → ln → Dirt-Samples → .tidal), with the
  crimewave/diams_dj/humpty stem→loop pairs as free ground truth
- README: export target corrected to Work/Sound/Samples + link step

The Sample Foundry: turns a track URL into separated stems ready for
sampling, replacing the manual yt-dlp → demucs → Audacity dance.

- engine/ — pure-ish, importable steps (model/fetch/separate); heavy work
  shells out to dedicated venvs so the engine runs under system python3
- separate.py = pluggable backend REGISTRY: demucs (engine1, live) +
  roformer (engine2, stubbed) behind one --backend interface
- foundry.py — CLI driver (fetch/sep/catch/list/backends/serve), tide.py-style
- server.py — stdlib GUI backend with job-progress + HTTP Range (audio seek)
- ui/index.html — self-contained GUI in the armada Ship's Bridge tokens;
  stems colour-coded by role (bass/drums/other/vocals)
- tests/ — 10 unit tests on the pure parts (slug, info-parse, argv, registry)
- README — workflow, engine1→engine2 roadmap incl. Roformer hyperparams +
  6GB VRAM caveat, phase-2 loop-finder charter, UX surface

Validated end-to-end on the Loituma test catch.

copyfile() threw BrokenPipeError/ConnectionResetError (Errno 104) when a
browser aborts/seeks a Range request mid-stream — normal for <audio>
scrubbing, but it spammed a traceback. Wrap both copy branches and close
the file handle in a finally. Verified: mid-stream abort leaves no
traceback and the server keeps serving (206 follow-up OK).

Adds the edit/remaster/pull/permanence axis the cost model was missing:

- 4th cost bucket UPDATE: re-mastering an already-live release. Free on
  RouteNote; a new paid release on CD Baby; forces DistroKid to the Ultimate
  tier ($89.99/yr) for an in-place Audio Swap. Driven by a Remasters/year slider.
- Freedom & control matrix: per-option remaster / edit-metadata / takedown /
  if-you-stop-paying, glyph+text+colour (never hue alone). Surfaces the key fact
  that no DSP distributor swaps audio in place (a store rule), with Bandcamp the
  free-replace outlier.
- Ownership reframed: all options are non-exclusive (you keep 100% of masters);
  what differs is permanence, edit-freedom, portability.
- Softened the unverified RouteNote freeze-risk claim; flagged Premium pricing
  as contested pending verification.

Both --extreme and --optimize disabled intel_pstate turbo on the theory
that frequency changes cause jitter. On HWP-capable CPUs (i7-10875H) the
performance governor already holds clocks steady at the top; disabling
turbo only caps all cores at the 2.3GHz base (~64% of clock headroom
thrown away), starving SuperDirt's DSP and causing the very xruns/crackle
it was meant to prevent. Now pins min/max_perf_pct=100 with turbo on.

Snapshot resolved tasks (foundation/distro, Ardour/boundaries, Judge UI, triangle
DRY, fleet colour, the sample-grounding fun-sprint, and the dataviz epic #64) so
the active board stays lean. Active board now = pending/in-progress only.

Field: per-cell DOMINANT family colour (was a mean of all overlapping hues →
brown mud); mixed/overlap zones now fade by 'purity' instead of muddying.
Tuned colourful-but-subtle behind the dots (desaturate 0.40 + slight lift to
white, maxα 0.44) — dots stay the stars.
Zoom/pan: wheel to zoom at cursor, drag to pan when zoomed, +/−/reset controls
(top-left) + double-click reset; dots grow with zoom; field & hit-testing follow
the view transform, clipped to the plot rect.

sample_families.json was built at a smaller per-folder cap → only 831 per-index
entries (825 of the 1485 plotted samples labelled). Re-ran the resolver at
--max-files 12 (same clap:fine method) over 150 folders → 1501 per-index entries,
1490 labelled. unwrapped.json now 100% family-coloured (was ~56%): no more grey
'unlabelled' dots, the density field reads everywhere.

Low-res IDW of dot colours (72x46 Gaussian-weighted, σ≈3.1), desaturated 50% and
faded by local density, bilinear-upscaled behind the dots → pink kick region,
blue keys region, purple bass / yellow hat pockets emerge as a slow gradient.
Cached offscreen, rebuilt only when colours/axes/filter change (never on hover);
'regions' toggle in the controls; auto-off in vibe-search mode.

Loading weights wasn't enough — the first forward still cost ~30s on torch's
one-time graph/thread init. Warm now runs a throwaway _embed_texts() so the first
USER query is ~1.5s, not 30s.

The first /vibe hit cold-loaded CLAP (~16s) with the UI stuck on 'searching…',
reading as hung. Now serve.py warms the model in a daemon thread at startup (when
semantics_embeds.npz is present) so the first query is instant; the client shows
'first search loads the model, ~15s' on a cold start and aborts after 90s with a
retry hint instead of spinning forever.

index.html ties the corpus dataviz together (By the numbers · Unwrapped · The
Triangle) and opens with a typographic style cloud sized by TRACK count — read
from PLN's own gig metadata (catalog_view metas[].style via models.norm_style),
never per-sample CLAP genre. dnb 18 · breaks 17 · techno 14 · nujazz 12 = the
TechnoJazz spine. build_landing.py bakes landing.json (cloud + corpus stats).
Ship's Bridge; magenta only on the live-dot + cloud hover.

- build_unwrapped: 2D PCA of the CLAP embeds → per-sample 'vibe map' coords +
  16 seed-vibe chips (PLN's own words — the on-ramp, since users don't know what
  to type). vibe0/vibe1 join the axis picker as a semantic-space lens.
- unwrapped.html: VIBE SEARCH box (free text → /vibe) + clickable seed chips;
  results highlight on the map (dim misses, size hits by similarity, ring the top,
  faint-violet→magenta ramp) and auto-reveal the vibe map; result strip auditions.
  Shift-click any dot → /similar (nearest neighbours in embedding space). Null-safe
  plotting for vibe/raw axes. Graceful banner when the endpoint is absent.

Katana-first finding: per-one-shot CLAP genre/mood tags are unreliable (every
hit → boom-bap/euphoric — a 0.3s sound has no genre), but the audio EMBEDDINGS
are gold for RELATIVE similarity. 'warm dusty rhodes' → suns_keys gold-keys +
west-coast electric; 'jazzy upright bass' → no_sunshine/come_bass loops; a kick's
nearest neighbours are other kicks (0.96 cross-folder). So we ship similarity,
not fake absolute labels (Principle 1: trust the instrument).

- sample_semantics.py validated on real audio; semantics_embeds.npz = 1490×512-d.
- serve.py: lazy CLAP /vibe?q= (embed any phrase → rank) + /similar?name= (by
  audio-embed cosine). 503 if unbuilt, 400/404 on bad input; static serving
  untouched. Single-user LAN, torch loads once on first hit.

ParVagues Unwrapped — a Ship's Bridge dark explorable over unwrapped.json:
- The map: canvas scatter of all 1,485 samples; pick any 2 of the 5 named
  superfeature axes (or raw features); colour by family / audio-cluster /
  folder-kind / folder-agreement; hover to identify, CLICK TO AUDITION the .wav.
- The five axes: PCA loading bars (what composes each superfeature) + RF
  family-discriminator importances (centroid + temporal-centroid on top).
- Family fingerprints: per-family beeswarm on a switchable feature — the
  kick-punchy / bass-sustained split made visible.
- Folders are loose: family x audio-cluster contingency grid (ARI 0.25) — the
  timbre-not-label story, the data behind 'a folder is a loose grouping'.
Zero npm deps; fleet family colours; magenta reserved for the audition pulse.

Joins per-file features × resolver family/kind × standardized PCA projection
(PC1-5) × KMeans timbral cluster × audition .wav path into one unwrapped.json
the browser explores without recomputing. Reuses feature_eda.load_matrix so the
matrix matches the MDA exactly. Names the 5 leading PCs as superfeature axes
(Brightness / Timbre / Loudness / Envelope kick-bass / Tonal-noisy), orienting
coords by a stable anchor so poles read right. Contract tests green.

PLN: each SAMPLE is classified individually; a folder is a loose grouping, often a
heterogeneous KIT. Added kind=single|dominant|kit; for kits folder_agrees=None and
the run shows KIT [fam+fam] not a misleading dominant+flag. Folder-name flag now fires
only for folders claiming ~one family (cpluck->synth still flagged). Regenerated palette.

sample_features.py overfetched 36 L0/L1 features × 1485 corpus samples; feature_eda
mines them three ways:
- correlation: only 2 redundant pairs ≥0.9 (duration~temporal_centroid 0.97,
  bandwidth~rolloff 0.91) → the overfetch was lean, 34/36 independent.
- PCA: intrinsic dim 19 (90%) / 24 (95%) — genuinely high-D. The 5 leading PCs are
  interpretable SUPERFEATURE AXES: PC1 brightness (rolloff/centroid), PC2 timbre
  (mfcc5-8), PC3 loudness (rms/peak/flux), PC4 envelope/time (temporal_centroid,
  decay_slope, attack — the kick↔bass axis), PC5 tonal-vs-noisy (kurtosis/chroma_entropy).
- clustering: KMeans(12) vs resolver families ARI=0.25 NMI=0.40 (timbral clusters
  partly orthogonal to semantic family — consistent with 'folders are loose').
  RF importance: spectral_centroid + temporal_centroid are the #1/#2 family
  discriminators → validates productizing the kick↔bass tiebreaker (#80).
TDD: 3 synthetic invariants (redundancy/dim/separation) + real-data load guard.

Layered resolver (L1 filename -> L3 batched CLAP audio -> L2 folder cross-check).
116 homogeneous, 33 kit-like, 20 folder-name disagreements flagged (cpluck->synth,
amencutup->snare, sunny_brass->pad, nujazz_keys125->lead...) — audio overrides the
name, flags surface for review not silent override. max_files=6, ~7min.

The ~88 fine descriptors were re-encoded through RoBERTa on every folder's
forward — a fixed cost that made batch size irrelevant. Now cache the normalized
text embeddings + logit scale at load; per-folder forwards run only the audio
tower (get_audio_features) and a matmul against the cached text embeds. ~1.8x
(14s->7.7s/folder; text was ~45% of per-folder cost).

API note (transformers 5.10.2): get_text/audio_features return a model-output
object whose .pooler_output IS the projected 512-d joint embedding — verified
identical to a full ClapModel forward to 1.6e-7. No classification change.

Opaque folders send every file through a batched HTSAT forward; --max-files N
caps that. Defaults to CLF.MAX_FILES (12). Use 6 for fast dominant-family
grounding when full per-index kit detail isn't needed.