docs(tasks): archive #79/#82/#78/#8/#61/#81 — the seed that never ran, and a…

docs(tasks): archive #79/#82/#78/#8/#61/#81 — the seed that never ran, and a task built on a guessed unit name

armada/tasks/completed-archive.md

@@ -832,3 +832,87 @@ refreshed on menu-open only. `is_running` rewritten forkless. +11 tests.
   room seconds before playing.
 
 **Deps.** Unblocks #25 (the Bridge as single cockpit).
+
+---
+
+## 2026-07-29 — the day the seed was found to have never run (OPAL D-5)
+
+### #79 — eval produces a SILENT pattern; the 4-cycle xfade masks it as a "drift to silence"
+**Description.** PLN, playing: *"ctrl_enter on do it right, i hear the 4-bar pattern
+each bar lower, 4th barely audible, its clearly a xfade"*, and on wap *"no bass? when
+i move the knob C5 it starts sounding"*. Filed as an xfade bug.
+
+**Done.** Root cause found and fixed in `2648074`. It was neither the xfade nor the
+tracks. `pulsar-tidalcycles` splits BootTidal.hs on BLANK LINES, strips the `:{`/`:}`
+it finds, and re-wraps each chunk in its own `:{ ... :}` — which holds exactly ONE
+statement. The #55 control seed was `:{ let _seed = ... :}` followed by `mapM_` and
+`putStrLn` with no blank lines, so all three fused into one statement and died with
+a parse error. **The seed never ran once between 2026-07-27 and 2026-07-29.** Fix =
+two blank lines, documented in-file as load-bearing. Verified live: the boot now
+prints `[BootTidal] seeded 51 LCXL controls (#55)` and PLN confirmed it.
+
+**Learnings.**
+- *Every `dN` is `xfade N` with `xfadeIn 4`.* The four bars you hear after an eval
+  are the PREVIOUS pattern leaving. A silent eval gets a graceful exit, so the bug
+  presents as a drift. **Sound after ctrl+enter is evidence the LAST eval worked.**
+- Measured with the new `silent-eval.py`: empty control map → **23 orbits silent
+  across 10 of 10 buildable tracks**, including `desire` d1-d6, the entire track.
+  With the seed → **zero**. Necessary and sufficient.
+- **Why every guard was green.** `check-boot.sh` proved the helper block typechecks,
+  the seed block typechecks, and the helpers emit events on an empty map. All true,
+  all irrelevant — they test the Haskell; the failure was in how the file is CHUNKED
+  AND FED. #61 had already suspected exactly this and was closed on that comfort.
+  *Verifying that code is correct is not verifying that it runs.*
+- Two hypotheses were killed from a FILE with no rig and nobody's ears, by the
+  gig-log `controls` view: DJ filters parked at 51/59/80 (open) and gMask/gMute all
+  at 0. That is the value of a parked-state readout.
+- Guard added (`check-boot-blocks.py`, check-boot pass 4): replays Pulsar's chunking
+  bug-for-bug — including that JS `.replace(str, str)` is NON-GLOBAL — and fails on
+  parse errors only. Negative-tested: names block 26 at BootTidal.hs:663.
+
+**Deps.** Blocked #77, #74, #13 — all now unblocked. Re-closed #61 with the real
+answer.
+
+### #82 — gig-log report: show each control's LAST VALUE
+**Done.** `d1b21c9`. `left at` column, a `controls` subcommand translating each
+parked value into what it MEANS, and `report --from/--to` windowing.
+**Learnings.** Windowing has one non-obvious requirement: xrun/throttle are RUNNING
+TOTALS, so a naive slice reports the whole session's count inside the window —
+wrong, and wrong high; `slice_session()` rebases. Transcribing `gDJF` for the Hz
+column surfaced that the lowpass half is LINEAR IN HERTZ, so the knob feels inert
+then collapses in its last 2% — one gesture producing both "went quiet" and "sounds
+lpf'd", the exact symptom pair #79 was filed with.
+**Deps.** The keystone that unlocked #79.
+
+### #78 — unmapped knobs must be DARK
+**Done.** `2acbd9c`. Published current-track file; `--map` publishes as well as
+paints; `--watch` follows it at 1 Hz and repaints.
+**Learnings.** The colour code was never wrong — `build_frame` already paints only
+bound CCs. The lie was WHICH BINDINGS it was handed: the watcher ran with no track,
+so it painted the convention board (40/40 lit) regardless of what was loaded.
+Measured: `do_it_right` binds 19/40, so **21 dead knobs were glowing**; wap 15;
+desire 18. **The value ramp did not create the problem — it made a pre-existing lie
+legible.** When a cosmetic change surfaces a complaint, check whether the cosmetics
+exposed bad DATA before changing the cosmetics back.
+**Deps.** Left #84 (publish from Pulsar's ctrl+enter path) as the honest remainder.
+
+### #8 — THERMAL: 17024 package-throttle events
+**Done.** Answered from the gig-log recording, no new work needed. **0 core and 0
+package throttles** across 10h48m with audio, 72 °C peak. Heat is exonerated; the
+17024 was an absolute counter including power excursions.
+**Learnings.** Counters are deltas, not totals — and *measure the set, not the
+corpus*: PLN corrected that only ~90 min of that session was playing, so every
+aggregate over the full 10h48m described the wrong thing. That correction is what
+motivated `report --from/--to`.
+
+### #81 — "gig-log recorder is DEAD"
+**Closed INVALID. The premise was false and the error was mine.**
+**Learnings.** I asked systemd about `parvagues-gig-log`, a name I GUESSED; the unit
+is `gig-log.service`, and it had been up 11 hours. A wrong unit name returns
+"inactive", indistinguishable from a dead service. I then used that false negative to
+resolve the report's deliberate ambiguity (*"still running, or it was killed"*) the
+wrong way — **a broken check settling the very question it could not answer**. Two
+lessons: read identifiers, don't guess them; and a tool must not offer the reader a
+choice it can make itself. `a4dc37b` makes the report decide from DATA RECENCY (not
+a process match, since `pgrep -f` has its own false-positive problem) and print
+either `● RECORDING NOW` or `the recorder was KILLED`.