docs(gig): structured archive for finished OPAL 2026 tasks

The /pre-compact ritual archives completed board tasks somewhere durable before
clearing them, so the learnings survive the task disappearing. armada/tasks/
completed-archive.md already does this — but its header scopes it explicitly to
L'Armada, the media/marketing toolbox. Gig prep is a different thread, and mixing
two documentary trails in one file makes both worse to mine later. Same discipline,
separate file.

Seven finished tasks captured in the established description/done/learnings/deps
shape: #15 runbook, #21 mute-bomb, #32 recorder, #19 disk archive, #9 launcher,
#27 highlight jitter, #5 HUD version control.

The learnings are the point, not the inventory. The ones worth re-reading:

- #21: a fix that lives only in a running process is not a fix. The gF rewrite was
  correct and live-validated for a full day while still being one SuperCollider
  reboot away from silently rearming. "Live-confirmed" and "durable" are different
  states and the gap between them is where this class of bug hides. The deeper
  design lesson: make rest position == no effect. The bug was never really the
  missing djfbus, it was that the hardware's DEFAULT state sent an EXTREME value.

- #9: order beats retry. The MIDI dance was a race symptom; gating on observable
  readiness (three independent checks, not a sleep) removed the race rather than
  automating the workaround. And idempotence is the property you want from anything
  you might run while panicking.

- #15: a frozen Pulsar does not stop the audio — patterns live in ghci, not the
  renderer — which inverts the instinct to panic-kill the editor.

- #19: du-deleted (436GB) vs df-freed (160GB) disagreed sharply; btrfs zstd + reflink
  sharing explains it, df is authoritative, don't chase the gap.

- #27: border participates in layout and reflows text, outline does not — for any
  decoration toggling at performance rates, outline is the correct primitive.

- #5: the HUD, pulsar-tidalcycles, and ~/.pulsar/styles.less are three separate
  version-control domains (the last untracked entirely, as user config). One visual
  bug can need coordinated edits across all three, which is exactly what #35 hit.

docs/opal2026-task-archive.md

+# OPAL 2026 — completed-task archive
+
+Long-term learning + blog/video source material for the gig-prep thread. Each entry is a
+finished board task captured in full (description · done · learnings · deps), appended by the
+`/pre-compact` ritual before the task is cleared off the active board.
+
+Deliberately **separate from `armada/tasks/completed-archive.md`**, which is scoped to
+L'Armada (the media/marketing toolbox). Same discipline, different thread. Newest first.
+
+---
+
+## #15 — Mid-set recovery runbook ("when it breaks")
+- **Description:** PLN needed a playbook for the gig going wrong on stage, where the failure
+  mode is panic as much as it is technical.
+- **Done:** `RUNBOOK.md` at repo root (commit `9bdf395`). Symptom-first structure, ordered
+  cheapest-first, with an explicit rule zero that a SuperCollider reboot is the *last* resort.
+- **Learnings:**
+  - Structure it **symptom-first, not cause-first** — mid-set you don't know the cause, you
+    know the sound stopped. §0 is a 10-second triage that routes to a section.
+  - **A frozen Pulsar does not stop the audio.** Patterns live in the ghci process, not the
+    renderer, so `Window: Reload` is a cheap fix that keeps the music playing. This inverts
+    the instinct to panic-kill the editor.
+  - Documented §2b, the **zombie port 6010** failure (see #21 learnings) — the nastiest kind,
+    because it is completely silent.
+  - **Fader 77 is global gain.** Down = everything silent, nothing looks broken. Now in the
+    pre-set checklist because it's a one-move fix that's easy to overlook.
+  - Every procedure is **labelled by evidence strength** — verified-by-running vs.
+    read-from-source vs. inherited-from-prior-incident. A runbook you've never executed is a
+    hypothesis and should admit it.
+- **Deps:** feeds #14 (rehearsal must actually *execute* the runbook, especially §5).
+
+## #21 — The gF/djfbus mute-bomb
+- **Description:** `gF1/2/3` sent values into `djfbus`, a **non-registered** SuperDirt effect.
+  LCXL faders rest at 0, so any gF-wrapped stream muted with no recovery — and essentially
+  every mature ParVagues track wraps its streams in gF. Armed across the whole set.
+- **Done:** Safe-at-rest rewrite using core `hpf`/`lpf` mapped so fader-rest == bypass
+  (`gHPF = range 20 8000`, `gLPF = range 20000 180`, `gF3 = id`). Shipped first as a
+  shadow-eval in `live/lib/prelude.tidal` (`29ce103`), live-confirmed by PLN ("f123 djfs seem
+  to work great overall"), then **baked permanently into `BootTidal.hs`** (`98d073f`).
+- **Learnings:**
+  - **A fix that lives only in a running process is not a fix.** It was correct and validated
+    for a full day while still being one SuperCollider reboot away from silently rearming.
+    "Live-confirmed" and "durable" are different states; the gap between them is where this
+    kind of bug hides.
+  - Design controls so **rest position == no effect**. The bug wasn't really the missing bus,
+    it was that the *default* state of the hardware was an *extreme* value. `gF3 = id` is the
+    strongest form of this — a spare fader that cannot touch sound at all.
+  - Discovered while type-checking: a stale `ghc` holding port **6010** makes a fresh Tidal
+    print "Can't handshake with SuperCollider" and then behave **completely normally in
+    Pulsar** while nothing plays and no error surfaces. Detect with `ss -lunp | grep 6010`.
+  - `prelude.tidal` was demoted from accidental-source-of-truth to explicit **rescue sheet**,
+    with the stale-boot tell written down (push fader 49 → mutes; pulling to 0 makes it
+    *worse*, because 0 is the mute position on the old defs and bypass on the new).
+- **Deps:** unblocked the same bake-in for #13; feeds #15's §1a and #18.
+
+## #32 — Always-on lean gig recorder
+- **Description:** Capture every performance without depending on Ardour, and slice test
+  samples for the Slop visuals work.
+- **Done:** `tools/gig_record.sh` (`29ce103`) — PipeWire → rotating hourly opus segments in
+  `recordings/`, with `start`/`stop`/`status`/`slice` subcommands. Verified running
+  continuously across this session (pid 3686091).
+- **Learnings:** deliberately a **separate process from Ardour** — if Ardour dies you still
+  have the set. A freshly-rotated segment reading 0 bytes is normal and should not read as
+  failure; `status` had to be taught to say so.
+- **Deps:** spawned #37 (GUI toggle, HUD REC indicator, silence-trim, per-track metadata).
+
+## #19 — Free 100GB+ before the gig
+- **Description:** Disk was at ~28GB free with a deadline of ~Aug 1; Ardour stem recordings
+  were the bulk.
+- **Done:** Freebox mount repaired, 436GB of **verified-identical** stem media archived and
+  deleted → **188GB free**.
+- **Learnings:** `du`-summed bytes deleted (436GB) vs `df`-reported bytes freed (160GB)
+  disagreed sharply. Cause is btrfs `compress=zstd` plus reflink/CoW sharing — benign
+  filesystem semantics, not partial deletion. **`df` is the authoritative number**; don't
+  chase the gap. Verify identity *before* deleting, always.
+- **Deps:** unblocked all recording-heavy work (#32, rehearsals).
+
+## #9 — 1-click ordered gig launcher
+- **Description:** At every gig start, whichever of SuperDirt or Ardour opened the
+  LaunchControl XL first won it. When Ardour won, sclang couldn't get the controller and PLN
+  was stuck doing a disconnect/reconnect MIDI dance during setup.
+- **Done:** `gig-up.sh` (`ba8e84a`, hardened by `e304355`, `0c65fda`, `2bc6a8e`, `c35c902`) —
+  brings SuperDirt up **first**, waits on a real readiness gate (scsynth alive + `:57120`
+  listening + a SuperCollider MIDI client present), and only then launches Ardour and Pulsar.
+- **Learnings:**
+  - **Order beats retry.** The MIDI dance was a symptom of a race; gating on readiness
+    removed the race instead of automating the workaround.
+  - Gate on **observable readiness**, not `sleep`. Three independent checks, because any one
+    of them alone can be true while the rig is still unusable.
+  - Made **idempotent** — it converges the rig rather than spawning duplicates, so it's safe
+    to run when you're unsure of the current state. That's exactly the property you want from
+    something you'd run while panicking (see #15 §5).
+  - `set -u` bit hard: `open_term` read an unused `$1` and crashed the launcher on start
+    (`e304355`).
+- **Deps:** feeds #15 §5, #18; prerequisite for #14.
+
+## #27 — Editor highlight horizontal jitter
+- **Description:** Eval highlights visibly shifted text horizontally when toggling.
+- **Done:** `border` → `outline` across `.event-highlight` and all 17 per-stream `-dN` rules.
+- **Learnings:** **`border` participates in layout and reflows text; `outline` does not.** For
+  any decoration that toggles at performance rates, outline is the correct primitive. (The
+  *residual* wiggle after this fix turned out to have a separate cause — see #35, a competing
+  runtime style injection toggling 30×/sec.)
+- **Deps:** partially overlaps #35; both feed #6.
+
+## #5 — Version-control the pulsar-parvagues-hud package
+- **Description:** The HUD package was live-edited with no git history.
+- **Done:** Package brought under version control with its own repo/branch.
+- **Learnings:** the HUD, `pulsar-tidalcycles`, and `~/.pulsar/styles.less` are **three
+  separate version-control domains** (the last untracked entirely, as user config). A single
+  visual bug can require coordinated edits across all three — which is exactly what #35 hit.
+  Knowing the domain boundaries is a prerequisite for debugging anything editor-side.
+- **Deps:** prerequisite for #6, #7, #22, #35, #36.