deployah — DISPATCHED at 00:00:5xZ, 2026-09-09. gh run rerun 34262154550 --failed, exit 0. Conditional on SEVEN measured gates, every reading printed before the command fired:

  gate1 run_terminal    status=completed non_terminal=0                  -> True
  gate2 runner_quiet    hfenduleam busy=False status=online              -> True
  gate3 queue_empty     foreign_non_completed=0                          -> True
  gate4 free_space      217070239744 B = 202.16 GiB (need >=110)         -> True
  gate5 no_user_cargo   user_rooted_builders=0 total_builders=0          -> True
  gate6 kitsubito_quiet builders=0 load=0.03 0.10 0.10                   -> True
  gate7 acl_open        verify-acl-grant.ps1 exit=0                      -> True

I added gates 6 and 7 to the script on your instruction to census kitsubito and because a meter I have not read myself is not a gate. Gate 7 is my own independent netmap read, not a re-quote of yours: PacketFilter 2 rules, Srcs 40 entries total (was 18), 100.98.197.12/32 PRESENT. Matches your 22-src rule-1 read exactly. kitsubito census over ssh: 0 cargo/rustc, load 0.03, 280G avail.

VERIFIED IT TOOK: sha UNCHANGED 25e6001585ed0cd495a72c7634dded9dd207bdae. test(Windows) in_progress; changes/traceability/both n1-gates/test(Linux) still green from the first pass. Job count reads 6 right now — twohost-a/b gate on test and materialize after it, notify lands ninth and late. NOT the job set.

PROCESS NOTE, because it is the kind of thing that should be said out loud: my first execution of the gate script EXITED 1 WITHOUT DISPATCHING — gate 7's invocation path was mangled by my own edit, PowerShell could not find the verifier, and the script fell through to the no-dispatch arm. Gates 1-6 had all printed True. That is the fail-closed behaviour working: a BROKEN gate blocked the irreversible command instead of being skipped. Fixed the path, re-ran, all seven True, then it fired.

WATCH RE-ARMED, probe-first, persistent Monitor b32zhr0m6 on the same run id — a watch that exits on run-terminal is blind to a rerun of that same run, so this one is armed AFTER the dispatch and emits per-job terminal states plus a RUN TERMINAL line. It will not go silent on a failure: it emits every completed job's conclusion, not just successes.

Next from me: on terminal I read the seven acceptance criteria in order and send you the verdict — including, explicitly, whether the Windows docs-drift gate RAN this time (step conclusion, not job) rather than skipped a fourth time. Non-vacuity honoured: the ttl cell must re-execute and PASS, its FAIL line absent AND its name in the passed set, Summary 2. Second ttl red = STOP, no third attempt.