docs(ci-reds): U12 --- gate run 1 was 15/16, red on the budget+PTY family

Head-exact gate at 6142acc: fifteen of sixteen stages green, including
sweep, m4, gpu, diff-check and all eight touched acceptance suites. The
red was 04-lib-crdt, where composition_overhead_under_ten_percent
(1.247x against a 1.10x budget) and
setsid_escapee_is_not_reaped_and_teardown_reclaims_readers failed
together. Both green on isolated rerun.

Filed as U12 rather than folded into U6 or U9, because both of those
instruct it: U6 says one of its selectors redding without the other is a
separate incident, and this is composition_overhead alone for the second
time; U9 is the same budget-plus-PTY shape but in 11-sweep with a
different PTY selector.

src/process.rs is not touched by this branch at all. src/editor.rs is,
but only in the panel-replay paths, not in composition.

The row does NOT claim load caused it. It records that the run was
knowingly taken on a machine that was quieter but not quiet --- load
11.04 at the start, 27.79 five-minute at the end, two foreign python
processes throughout, an apt install shortly before --- which are
conditions, not a mechanism.

Four incidents in this family now, and the discriminating control U9
named remains unrun: pin test-binary concurrency to 1, and separately
load a lone --lib binary.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016bqGA6s9tTUFzYpbeW3tai

docs/active-work.md

@@ -406,7 +406,17 @@ from #171 and #215 — the correction the 1b lane missed, honoured here.
       gone, so the completion has nothing left to clear and the
       gesture ending is the whole effect. Written into the row rather
       than left as a silently absent assertion.
-  - **REMAINING: the full head-exact gate, then the PR.** The gate
+  - **GATE RUN 1, head-exact at `6142acc`: 15/16, red on `04-lib-crdt`
+    only.** `composition_overhead_under_ten_percent` (1.247× against
+    1.10×) and `setsid_escapee_is_not_reaped_and_teardown_reclaims_readers`
+    failed together; both green on isolated rerun. Recorded as **U12**.
+    `src/process.rs` is not touched by this branch at all. **The run
+    was knowingly taken on a machine that was quieter but not quiet** —
+    load 11.04 at start, 27.79 five-minute at end, two foreign `python`
+    processes throughout — so it is reported as a red under stated
+    conditions rather than dismissed.
+  - **REMAINING: a head-exact gate on a genuinely quiet machine, then
+    the PR.** The gate
     wants a QUIET machine: a foreign C++/java build has been running at
     load 114+ through this work, and the wall-clock rows
     (`composition_overhead_under_ten_percent`,

docs/ci-red-signatures.md

@@ -963,3 +963,31 @@ tree.
 the measured `duration_ms` in the assertion message — is **owed its own
 small lane**. It was deliberately kept out of #242, whose diff does not
 touch `src/async_runtime.rs`.
+
+### U12 — U9's shape in `04-lib-crdt`: a budget test and a PTY test, together
+
+Recorded during the panel-replay lane's gate, 2026-08-21, on head
+`6142acc`. **Filed rather than folded into U6 or U9**, because both of
+those tell it to be: U6 says one of its selectors redding without the
+other is a different incident, and this is `composition_overhead_under_ten_percent`
+alone again; U9 is that same shape but in `11-sweep` with a different
+PTY selector.
+
+| field | value |
+| --- | --- |
+| **selectors** | `--lib --features crdt editor::tests::composition_overhead_under_ten_percent` **and** `process::tests::setsid_escapee_is_not_reaped_and_teardown_reclaims_readers`, failing in the same step |
+| **job / step** | local (Linux), `scripts/gate` step `04-lib-crdt`, log `20260821T122852Z-2922631` |
+| **required fragments** | `composition machinery added more than 10% overhead` / `live runtime probe` |
+| **observed** | **1.247×** against the 1.10× budget (single 192493 ns, dispatch 240130 ns); the PTY row failed at `src/process.rs:5155`, where `active_reader_probe` found no live reader within the 2s `Started` window |
+| **status** | **one occurrence, both selectors green on isolated rerun** |
+| **the rest of the run** | **15 of 16 stages green**, including `sweep`, `m4`, `gpu`, `diff-check` and all eight touched acceptance suites |
+| **what IS established** | both are timing-dependent by construction — one a wall-clock ratio, the other a 2-second liveness window — and each passed alone immediately afterwards. **`src/process.rs` is NOT touched by the observing branch at all**; `src/editor.rs` is, but only in the panel-replay paths, not in composition |
+| **what is NOT** | that load caused it. Load was **11.04 at the gate's start and 27.79 (5-minute) at its end**, with two foreign `python` processes at ~2 cores throughout and an `apt`/`dpkg` install shortly before. Those are conditions, not a measurement of the mechanism, and the run was **knowingly taken on a machine that was quieter but not quiet** |
+| **relation to U6** | its selector, alone again, in U6's own step. U6's instruction to judge that separately is honoured for the second time — see U9, which did the same |
+| **relation to U9** | the same budget-plus-PTY co-failure, in `04-lib-crdt` rather than `11-sweep`, with `setsid_escapee…` where U9 had `m6_1_pty_raw_mode…` |
+
+**This family has now produced U6, U9, U10 and U12, and the
+discriminating control U9 named is STILL UNRUN**: pin test-binary
+concurrency to 1, and separately load a lone `--lib` binary. Four
+incidents is enough evidence that the family will keep costing review
+rounds until someone runs it.