todlando -> doyle. PACER CHECKPOINT, 20:38:14Z. Nothing interrupted; evidence read off disk while the driver runs.

a-2: COMPLETE. serve_rc=0 stop_rc=0 wall_ms=1890 outcome=completed, row bootstrap-firewall leg=verify-query program=powershell.exe. Binder VERDICT SAME.
a-3: ACTIVE. Binder VERDICT SAME already recorded; the trial row has not landed yet, so its CPU bracket is still capturing. This is the command I am not interrupting.

SETUP: PENDING, and specifically NOT blocked. Zero request files exist on disk -- measured with find over the run root, not recalled. No nonce has been minted, no clock is running, and liam has been asked for nothing. Setup cannot be requested until a-3 closes and Arm A reconciles.

Binder comparison is now 3 for 3 SAME across a-1, a-2 and a-3, each against the same backslash-vs-forward-slash pair. The repaired comparator is behaving consistently on the real capture rather than passing once.

Arm A timings so far: 2491 ms and 1890 ms, both outcome=completed with a stop_rc=0 per arm. I am not drawing a timing conclusion from two rows, and per your earlier ruling these are UNPOPULATED-arm measurements -- they do not answer populated-arm timing and they do not establish admission.

No reconciliation verdict and no pair census have been emitted yet; both are Arm A's closing steps and both are what I require before Arm B, as you ruled.

States live for cleanup: pool_acquired, listener_may_exist, daemon_may_run. Still NO fw_mutation_may_have_occurred -- correct, nothing elevated has been requested.

liam last re-confirmed 20:29:11Z (4UU6KSP2). I re-ping immediately before the setup request regardless, since an ACK is a current-availability check and nine minutes is not now. I retain supervision of any outstanding handoff through cleanup. Next report: setup result or unexpected failure.