Metal judges read the channel their line crosses on

issues/audio/hda-client-stall-reads-one-resume-where-its-judge-wants-two.md

@@ -0,0 +1,50 @@
+---
+status: open
+kind: defect
+opened: 2026-09-29
+---
+
+# `hda_client_stall` reads one resume on the T14 where its judge wants two
+
+The T14 run of `wt/toyos-metaljudges` at `6737a442` reds the row with:
+
+```
+FAIL hda_client_stall: soundd resumed 0 time(s) — the second stream did not find a suspended daemon, so nothing here tests a resume:
+```
+
+The job itself exits 0 and prints `stalled 8 then 2 times, soundd survived`.
+Its cause is not measured; whether the judge or soundd is wrong is open.
+
+## Measured
+
+The `testcases` boot's log, from the job's spawn to its end, in order:
+
+```
+{… 6.275 soundd} soundd: client 0 removed (closed)
+[… 6.276 cpu6] spawn: /system/bin/test_rs_hda_client_stall pid=9 …
+{… 6.276 soundd} soundd: wakes=662 … clients=0 …
+{… 6.333 tid=1 soundd} soundd: opening stream: 44100Hz 2ch fmt=0
+{… 6.333 soundd} soundd: client 0 connected (id=1)
+{… 8.565 soundd} soundd: client 1 removed (closed)
+{… 8.565 soundd} soundd: wakes=126 … clients=0 …
+{… 8.585 soundd} soundd: suspended
+{… 8.836 tid=1 soundd} soundd: opening stream: 44100Hz 2ch fmt=0
+{… 8.836 soundd} soundd: client 0 connected (id=2)
+{… 8.836 soundd} soundd: resumed
+{… 9.755 soundd} soundd: wakes=480 … clients=0 …
+{… 9.780 soundd} soundd: suspended
+```
+
+- The window `tests/common/audio.rs`'s `job_window` cut for two sessions ends
+  at the `8.565` stats line: the `clients=0` stats line at `6.276`, after the
+  spawn and before the first stream opens, is counted as the first session's
+  end. The printed window carries no `soundd: resumed`.
+- Across the job's whole span the log carries one `soundd: resumed`, at the
+  second stream. The first stream opens with no `soundd: suspended` between
+  the previous client's removal at `6.275` and its open at `6.333`, so
+  `client_stall_on_metal`'s `resumes < 2` reds on that span too.
+
+## Exit condition
+
+`hda_client_stall` PASS on the orchestrator's T14 run of the head that lands
+the fix, and this file is deleted.

issues/audio/hda-tone-reads-underruns-on-the-t14-where-its-judge-wants-none.md

@@ -0,0 +1,50 @@
+---
+status: open
+kind: defect
+opened: 2026-09-29
+---
+
+# `hda_tone` reads underruns on the T14 where its judge wants none
+
+The T14 run of `wt/toyos-metaljudges` at `f5c80264f` reds the row with:
+
+```
+FAIL hda_tone: soundd filled 104 period(s) the tone client had not covered, on a client that keeps its ring full:
+```
+
+The job itself exits 0 and prints `tone done`. Its cause is not measured;
+whether the judge or soundd is wrong is open.
+
+The T14 run of `main` at `7e151819` reddened the row earlier, on `the
+testcases's log carried no kernel output at all`.
+
+## Measured
+
+The `testcases` boot's log over the window the judge printed, selected lines in
+order:
+
+```
+{… 3.021 tid=1 soundd} soundd: opening stream: 44100Hz 2ch fmt=0
+{… 3.021 soundd} soundd: client 0 connected (id=0)
+{… 3.021 soundd} soundd: resumed
+{… 5.024 soundd} soundd: wakes=984 completions=689 submitted=697 underruns=0 drains=0 … clients=1 …
+[… 6.252 cpu0 tid=1] exit: test_rs_audio_tone tid=1 code=0 cpu=12ms (after 48 like it suppressed)
+{… 6.252 pid=8 test-runner} tone done
+{… 6.252 test-runner} ===TEST_END test_rs_audio_tone exit=0===
+{… 6.252 test-runner} ===TEST_START test_rs_hda_client_stall===
+[… 6.253 cpu7] spawn: /system/bin/test_rs_hda_client_stall pid=9 …
+{… 6.254 tid=1 soundd} soundd: opening stream: 44100Hz 2ch fmt=0
+{… 6.254 soundd} soundd: client 1 connected (id=1)
+{… 6.258 soundd} soundd: client 0 removed (closed)
+{… 7.024 soundd} soundd: wakes=989 completions=688 submitted=688 underruns=39 drains=0 … clients=1 …
+{… 8.412 soundd} soundd: client 1 removed (closed)
+{… 8.418 soundd} soundd: wakes=741 completions=479 submitted=479 underruns=65 drains=0 … clients=0 …
+```
+
+The printed window carries three stats lines, with `underruns=0`, `underruns=39`
+and `underruns=65`, which sum to the 104 the row names.
+
+## Exit condition
+
+`hda_tone` PASS on the orchestrator's T14 run of the head that lands the fix,
+and this file is deleted.

issues/build/a-metal-only-row-cannot-be-disabled.md

@@ -16,8 +16,7 @@ row before any boot.
 
 ## Measured
 
-`hda_tone` and `hda_client_stall` red on the T14 run of `main` at `7e151819`
-(`issues/build/a-readbacks-kernel-records-never-count-as-kernel-output.md`).
+`hda_tone` and `hda_client_stall` red on the T14 run of `main` at `7e151819`.
 Only the code reading is verified: `check_redlist` tests `schedule.contains`, and
 the metal branch filters `METAL` rows through the same `keep`. Running
 `--known-red hda_tone` and `--list` against a tree with such a row is owed and

issues/build/allocator-stress-bounds-a-failed-reservation-by-sixteen-gib.md

@@ -0,0 +1,20 @@
+---
+status: open
+kind: tooling
+opened: 2026-09-29
+---
+
+# `allocator_stress` expects a 16 GiB reservation to fail, which a machine with more memory grants
+
+`test_oom_graceful` (`tests/toyos-rust-tests/src/bin/allocator_stress.rs`)
+asserts `try_reserve(16 * 1024 * 1024 * 1024)` is refused as "way more than
+available RAM". That is a guest's size, as the total-memory range beside it
+was: the shared metal boot runs the same binary on the T14, which reports
+`mem_total=16777216000` (main's T14 run at `7e151819`, boot `shared`), 384 MiB
+short of the reservation. A machine with more than 16 GiB free grants it and
+reds the test.
+
+## Exit condition
+
+The refused size derives from the total `sysinfo` reports rather than from a
+constant.