log 047: the same bug twice in one morning, writer changed and reader lied

Coda on the reader patches, plus the measurement worth keeping: scsynth sits at oom_score_adj=200, penalised rather than merely unshielded, which is the real argument for getting protect back on.

armada/tasks/047-the-fix-that-warmed-nothing.md

@@ -74,3 +74,37 @@ it will drift, and the drift shows up as false confidence rather than as an erro
 
 "I fixed the preload and it went from 132 sample banks to zero. Then the checker
 told me everything was fine — and *that* was the real bug."
+
+---
+
+## Coda, same morning: the identical bug one file over
+
+While reviewing the `fix-ardour-rt.sh` dry run before applying it, the same shape
+turned up in the Ardour work. Four patches correctly stopped **both writers** from
+promoting Ardour's GUI thread to `SCHED_FIFO/80`. Nothing patched the **readers**:
+
+- `parvagues-protect --check` demanded `SCHED_FIFO` of every target, so an Ardour
+  deliberately left on `SCHED_OTHER` would have read `UNPROTECTED` forever, and
+  `check()`'s exit code is what the gate, the Bridge and gig-up believe. A correct
+  rig would have shown a red Ardour all night, burying the one question worth
+  asking: *is scsynth actually covered?*
+- `perf-audio` kept printing `✓ Set Ardour (PID …) to real-time priority 80`, and
+  the same for every child, while deliberately doing neither.
+
+Fixed in `06352a9`: `check()` reads the target's own prio out of its `TARGETS` row,
+so prio 0 means "want the OOM shield, leave scheduling alone". Verified on copies —
+8/8 verdict truth table, both files `bash -n` clean, second `--apply` idempotent —
+and then against the live processes, where the patched checker prints
+`ok ardour[615285] oom_score_adj=-1000 sched=SCHED_OTHER (oom shield only, by
+design)` and still flags scsynth and sclang, which are genuinely uncovered while
+protect is off.
+
+**Twice in one morning, in unrelated code: the writer changed and the reader kept
+asserting the old truth.** The preload version invented an all-clear; this one would
+have invented a fault. Both are the same defect wearing opposite clothes, and both
+were found by reading the thing that judges the change rather than the change.
+
+One measurement worth keeping from that run: **scsynth is sitting at
+`oom_score_adj=200`**. Not merely unshielded — *penalised*, i.e. first in line for
+the OOM killer on the whole box. That is the real argument for getting
+`parvagues-protect` back on, and it is invisible unless something asks.