docs: board + archive — the Fred pack landed, #20b and #21 written up

TODO.md: the Fred section moves from "RESUME HERE" to shipped, with the gate numbers,
the six-cuts-five-discarded history, and the audition-first list of 6 kitcheck
advisories that are now waiting on PLN's ears rather than on any tool.

completed-archive.md: #21 (La Cale) and #20b (the pack) as standalone entries. The
learnings worth carrying: the rack had to BE `loopAt` rather than resemble it; a gate
blocks on defects and reports properties, because blocking on a property is a gate
everybody learns to ignore; `n` is the folder's index, never the manifest's; a rounded
tempo is a stretched loop; and three tests that passed while proving nothing, which was
most of the day's actual value.

armada/tasks/completed-archive.md

@@ -1703,3 +1703,114 @@ measuring the output, never by reading code.
 **Deps.** Unblocked #21 (the auditioner SPA PLN asked for), #22 (same driver at the
 acapellas + the 27 demucs catches), #23 (kitcheck/restage over the 163 existing kits).
 Runs 1 and 2 were discarded — each predated a fix; run 3 is the deliverable.
+
+---
+
+### #21 — La Cale, the kit auditioner (browse · loop · rack · export)
+
+**Description.** PLN, after the first Fred kits landed: *"can you spawn a reusable SPA so
+i can listen to the samples, try each as a loop, quickly sequence? TBH could be a strudel
+playground i'm open to this idea lets think :P"*, then *"the SPA allows click-based
+looping not just strudel coding"*. The need behind it: a grade in a manifest is a claim,
+and the claim had been wrong twice in a week. Nothing in the toolbox could turn 150 cut
+loops into a decision.
+
+**Done.** `armada/ui/kits.html` + `src/kits/{engine,tidal,types,Waveform,SampleRow,Rack,
+KitAuditioner}` + `tools/foundry/kitindex.py`. Merged to master as part of `0f27184`;
+built across `ec20026`, `18ec7b4`, `e2b701e`, `c9e7558`, `1edeecf`, `ee8d4ac`.
+Browse every kit under the sample root (the 111 pre-Foundry ones included), audition a
+sample untouched or looped, rack several on one bar grid at a chosen tempo, export a
+paste-ready `.tidal` block. 24 vitest cases; `npm test` added to the project.
+
+**Learnings.**
+
+- **The rack had to BE `loopAt`, not resemble it.** `rate = dur / (bars × 4 × 60/bpm)` is
+  exactly what `loopAt bars` does, so what the browser plays is what the rig plays and the
+  export is the same two numbers in Tidal's syntax. Anything looser would make the tool a
+  preview instead of a test.
+- **Scheduling per bar against an absolute grid, not free-running nodes.** The cutter's
+  bar tolerance is 1 ms, which is 60 ms of skew after a minute — enough to hear.
+- **Two audition modes because they answer different questions.** Once = the file
+  untouched (attack, tail). Loop = the seam, which is the only thing that decides
+  usability, because it is the sound the audience gets every bar.
+- **Deliberately not wavesurfer**, though the project already depends on it: a hundred
+  rows would mean a hundred fetches and decodes to draw 200 px. `kitindex.py` ships a
+  128-point envelope per sample; audio is fetched only on play.
+- **Strudel was considered and rejected as the play engine.** It is Tidal syntax in the
+  browser, but the deck already plays the rack and a second audio graph is a second thing
+  that can disagree about what the same cells mean. The tab emits text for strudel.cc.
+  The dependency was uninstalled — one nothing imports is a false statement about the
+  project.
+- **The fader is linear, Tidal's `gain` is quartic.** 0.5 exports as `gain 0.84`. Copying
+  it across is a 12 dB error. Pinned by a test.
+- **Three rounds of "screenshot it rather than assume it renders"** found what typecheck
+  and lint could not: truncated filenames with no tooltip, a dead `! 0` control, a kit
+  tempo rounded 132.5→133 (which `loopAt` turns into a 0.4% stretch of every loop in the
+  kit), three loops off one stem labelled identically (fixed by surfacing `start_s` as
+  `@4:31`), and CLAP tag chips painted over the first 15% of every waveform — i.e. over
+  the downbeat, which read as a gap in the audio and nearly sent me hunting a bug in the
+  envelope generator.
+- **It cannot be a published Artifact**: the CSP cannot reach a localhost audio server.
+- Two staleness traps closed: the envelope cache keys on path+size+mtime (a restaged kit
+  must miss it — a stale envelope over fresh audio is a lie that looks like data), and the
+  index is written to `dist/` as well as `public/`, since `npm run build` copies `public/`
+  exactly once.
+
+**Deps.** Needed #20's manifests to have anything to browse; unblocks #23 (auditing the
+111 pre-Foundry kits) by making a verdict cheap to reach.
+
+---
+
+### #20b — The Fred pack shipped (150 samples, 43 kits)
+
+**Description.** Turn Fred again..'s dropped stem dropboxes into loadable Tidal kits, one
+per track × role family, and have PLN able to validate them by loading them.
+
+**Done.** Merged to master (`0f27184`, `831d3f1`, `ebcd566`). 150 samples · 43 kits · 9
+tracks at `~/Work/Sound/Samples/Fred/output/<kit>/`, symlinked into Dirt-Samples,
+registered `third_party` with provenance. Record in `tools/foundry/packs/fred/`. Python
+suite 112 → 128.
+
+**Learnings.**
+
+- **Silence was the grader's optimum.** 24 of 153 loops were digital silence at −104…−75
+  dBFS, graded 0.707–0.896, tier A and B. Every term in the loop rubric is vacuously
+  perfect on nothing — the seam between two silences is immaculate, the DC is zero, the
+  zero crossings trivial, the length bar-exact by construction — and level carried 10%.
+  Nobody wrote "prefer silence"; it fell out of five correct measures, none of which had
+  an opinion about whether there was audio. **Presence is a precondition, not a weighted
+  term.** The floor was PLACED in a measured 28.4 dB gap, not tuned.
+- **Rejecting late is not the same as rejecting early.** Filtering at export took BIGHEN's
+  tonal kit from four silent loops to *one* real one; gating before the finder ranks took
+  it to two, and brought its fx kit back from nothing. A rejected slot is a slot spent.
+- **A relative check cannot see a uniform failure.** The dead-bar test compares bars to
+  each other, so it catches a fade-out and is blind to a clip that was never there.
+- **The gate blocks on defects, reports properties.** It first blocked all 150 samples on
+  20 `near-silent` and 17 `mono-incompatible` files, none broken: a kit is gain-staged as
+  a unit and never boosted, so its quiet members are quiet *on purpose*, and low
+  interchannel correlation in a vox double is the width someone paid for. Blocking on a
+  property is not the cautious direction — it is a gate nobody can pass, which is a gate
+  everybody learns to ignore. `mono-incompatible` **does** block on bass, where mono
+  collapse kills the fundamental and nobody chose that.
+- **`n` is the folder's index, not the manifest's.** Export writes into a kit dir it never
+  empties, so a re-cut that renames one file leaves both and shifts every index after it —
+  silently, with a correct-looking manifest and cheat sheet. kitgate now compares the two
+  listings index-for-index.
+- **A rounded tempo is a stretched loop.** The cheat sheet printed "133 bpm" over files cut
+  at 132.51. Each kit now prints its own `setcps`, because `loopAt` squeezes into N bars of
+  the *current* cps and leaving the ÷60÷4 to the reader is what makes the rounding bite.
+- **Restage was not idempotent**: PCM_24 quantization left the peak a hair above the
+  ceiling, so it asked to restage forever, requantizing every file each round. 0.05 dB
+  dead band.
+- **A background job that takes the desktop down is not a background job.** Nine children
+  at 88% of a core each → load 15.6/16 on the machine PLN performs on, and he said so
+  mid-run. Half the cores by default, `nice 19` + `SCHED_IDLE` via a preexec hook.
+- **Three tests passed while proving nothing**, and finding that out was most of the day's
+  value: wrong fixture (a stem gated on its own beat grid cannot generate a window inside
+  its own silence), wrong scope (a hook looked up via `__globals__` where a nested
+  function does not live → `None` → no-op lambda), wrong lens (a "sparse" fixture read
+  −32 dB RMS against a −45 threshold). The habit is now explicit: break the fix, watch the
+  test fail, put it back.
+
+**Deps.** Built on #20's `engine/stempack.py`; produced the data #21 browses. Unblocks #22
+(the acapellas, instrumentals and ~27 demucs catches) with no new code.

tools/foundry/TODO.md

@@ -56,46 +56,47 @@ python3 -m engine.stempack "<pack>" --jobs 9 --no-link --out /tmp/cut
 python3 kitgate.py /tmp/cut/fred_kits.json            # then --link
 ```
 
-## 🟡 In progress — finish the Fred pack (RESUME HERE)
+## 🟢 Landed (2026-08-22) — the Fred pack SHIPPED
 
-**Staging root is `~/Work/Sound/Samples/Fred/output/<kit>/`** (PLN asked for it there,
-2026-08-21), passed as `--samples-root`; Dirt-Samples symlinks point into it.
+150 samples · 43 kits · 9 tracks, staged at `~/Work/Sound/Samples/Fred/output/<kit>/`
+(PLN's choice) and symlinked into Dirt-Samples. `s "fred_marea_drums" # n 3` resolves.
+Record: `packs/fred/{fred_kits.json,.md,.tidal,kitcheck.md}`. Merged to `master`
+(`0f27184`); branch `origin/claude/fred-kits` kept as history.
 
-Runs 1-5 were all discarded, each superseded by a fix found by looking at its output:
-161 samples (pre-dead-bar-fix), 153 (pre-rotation-fix), 153 (**24 of them digital
-silence graded A/B**), aborted (wrong staging root), 129 (silence filtered at export
-only, so BIGHEN's tonal kit lost 3 loops instead of gaining 3 real ones). **Run 6** is
-the first with the finder itself refusing silent windows.
+    S×84 A×28 B×38 · 111 loops + 39 chops · 150 of 150 distinct sounds
+    restage 0/43 · 0 grade mismatches · kitcheck 0 failing · n indices verified
+    bank-check: "No shadowed bank names" · rights 43/43 third_party w/ provenance
+
+Ground truth held: MAREA declares `123BPM`, grid read **123.05**.
+PLEASEMAKEITBETTER at 20% tempo confidence correctly shipped chops only.
+
+**Six cuts, five discarded, each by looking at the output** — 161 (dead bars), 153
+(rotation), 153 (**24 files were digital silence graded A/B**), aborted (staging root),
+129 (silence filtered at export only, so BIGHEN's tonal kit lost 3 loops instead of
+gaining 3 real ones). Full account: `armada/tasks/028-la-cale-and-the-empty-loop.md`,
+memory `feedback_presence_is_a_precondition`.
+
+**Audition first** (6 kitcheck advisories, weak bar periodicity — bar-aligned but not
+repeating units): `marea_drums/10_marea_4b` and `11_marea_4b` (0.12 / 0.08 vs a 0.30
+drums floor — the suspicious pair), then `angie_tonal/03_padstuff_2b`,
+`angieinterlude_bass/00_pad_2b`, `bighen_tonal/01_synth_2b`, `yasminah_tonal/00_keys_4b`.
+Plus 20 quiet / 17 wide / 2 marginal seams, reported not blocked. **Awaiting PLN's ears.**
+
+To re-cut or cut another pack:
 
 ```bash
 cd tools/foundry
-python3 -u -m engine.stempack "$HOME/Work/Sound/Samples/Fred/fred again stems" \
-    --no-link --samples-root "$HOME/Work/Sound/Samples/Fred/output" --out <fresh dir>
+python3 -u -m engine.stempack "<pack dir>" --no-link \
+    --samples-root "$HOME/Work/Sound/Samples/<pack>/output" --out <fresh dir>
+python3 restage.py --cuts <out>/fred_kits.json --dry-run    # expect 0
+python3 kitgate.py <out>/fred_kits.json                     # 5 checks
+python3 ../../armada/tide-table/register_fred_rights.py <out>/fred_kits.json
+python3 kitgate.py <out>/fred_kits.json --link
+python3 kitindex.py --root <staging root> --root "$HOME/Work/Sound/Samples" \
+    --cuts <out>/fred_kits.json                             # the index La Cale reads
 ```
 
-`--jobs` now defaults to half the cores (children run at nice 19 / SCHED_IDLE — PLN
-works on this machine). Takes ~20-30 min, fully deterministic.
-
-**Exact steps to finish, in order:**
-1. `python3 restage.py --cuts <out>/fred_kits.json --dry-run` — expect **0 kits**
-   changed (export stages, and restage has a 0.05 dB dead band so it is idempotent).
-2. `python3 kitgate.py <out>/fred_kits.json` — five checks, must be green on grade /
-   kitcheck / shadowing / `n` indices. `rights` is advisory and WILL warn until step 3.
-3. `python3 ../../armada/tide-table/register_fred_rights.py <out>/fred_kits.json`
-   — every `fred_*` bank as `third_party` **with provenance**, so per-track release
-   blocks by design. Before linking.
-4. `python3 kitgate.py <out>/fred_kits.json --link` (takes each kit's root from the
-   manifest), then `python3 ../bank-check.py --since 1` to confirm 0 shadowed.
-5. `python3 kitindex.py --root "$HOME/Work/Sound/Samples/Fred/output" --root \
-   "$HOME/Work/Sound/Samples" --cuts <out>/fred_kits.json` — the index La Cale reads.
-6. Copy `fred_kits.{json,md,tidal}` + `kitcheck.md` into `packs/fred/` and commit
-   (`packs/README.md` — manifest, not audio).
-7. Report: per-kit tier histogram, distinct-sound count, the "worth a second listen"
-   table, and the paste-ready `n` indices.
-
-**Ground truth:** MAREA declares `123BPM` in its filenames; the grid must read
-**123.0**. PLEASEMAKEITBETTER reads 20% tempo confidence and correctly ships
-chops only.
+`--jobs` defaults to half the cores; children run at nice 19 / SCHED_IDLE.
 
 ## 🟢 Landed (2026-08-21) — #21 La Cale, the kit auditioner