todlando -> doyle. END. Attempt 2 complete 20:45:13Z, driver exit 3. ONE residual needs your decision; everything else is closed and verified.

=== VERDICT: SETUP NOT ACCEPTED, face unverified-after-write, pair_written=yes ===
original_rc=3, owned by the first failure as designed. Face counts, the driver's own census and not liam's word:
  reconciled=0  unverified=1  unverified_after_write=1  enforcement=1
  mismatch_written=0  source_store=0  no_lan_scope=0  mutation_disabled=0  clean=0
  cleanup_unverified=0  already_up=0  repair_requested=0  elevation_unavailable=0
Exactly one face lit, and it is enforcement. Your framing holds: the product reached an ENFORCEMENT rejection rather than attempt 1's spec-match rejection. That is a different failure, reached further in. It is NOT effective network admission and NOT a diagnosed cause, and I am not offering it as either.

=== ARM A, COMPLETE TABLE (unpopulated, pair ABSENT precondition retained) ===
  a-1  serve_rc=0  stop_rc=0  wall_ms=2491  outcome=completed  binder SAME  GUARD_UNCHANGED a-1-post
  a-2  serve_rc=0  stop_rc=0  wall_ms=1890  outcome=completed  binder SAME  GUARD_UNCHANGED a-2-post
  a-3  serve_rc=0  stop_rc=0  wall_ms=1863  outcome=completed  binder SAME  GUARD_UNCHANGED a-3-post
  Every row: bootstrap-firewall leg=verify-query program=powershell.exe.
  Binder 3/3 SAME on the real capture -- the r2 comparator resolving the backslash form against the forward-slash form. That comparison is what voided attempt 1 as a wrong-binder false negative; it is closed in the field.
  I am NOT reading these as admission: serveverb.rs:233-249 returns 0 unconditionally on LanUp. Only a nonzero is a real control error. Populated-arm timing remains UNANSWERED -- Arm B was held, so nothing here speaks to it.

=== SETUP CAPTURES (001, elevated, liam, executed ONCE, exit=0, 20:41:10Z->20:41:18Z) ===
  legs: verify-query 2581ms completed / verify-query 1485ms completed / reconcile-write 1864ms completed / verify-query 1892ms completed
  stdout: LAN_BOOTSTRAP_UP (256 bytes)
  stderr: LAN_FIREWALL_UNVERIFIED -- "The admission pair was WRITTEN and then could not be verified: Bootstrap rule spt-core-bootstrap-inbound-tcp is configured but ActiveStore enforcement is ["ProfileInactive", "NoLocalUser"], not Full." The product adds that this is NOT a refused write and suggests rerunning bootstrap. Not rerun: one nonce, one execution.
  Receipt 001 is properly formed and self-attesting: liam verified the exe sha independently (edd3d8e0, match), recorded elevation at 20:40:46Z, ran the run_exactly line verbatim with SPT_HOME explicit and SPT_INSTALL_NO_FIREWALL unset, and recorded pre-state (nothing on 29470 or 5470) and post-state (29470 LISTENING pid 52608, nothing on 5470).

=== D1 DIAGNOSTIC, PRESERVED SEPARATELY ===
d1-setup-rejection-snapshot.json plus d1-setup-rejection-render.out, captured BEFORE any teardown, as a SEPARATE invocation. Its stated ceiling, which I am passing on rather than burying: it bounds the host spellings present AROUND the call and keeps nothing else, and it CANNOT discriminate this candidate from the registration sha -- the QUERY const is byte-identical at 53d625cd and 921aa68f, measured.

=== TEARDOWN (002, elevated, liam, executed ONCE, exit=0, 20:44:11Z->20:44:15Z) — VERIFIED, not merely claimed ===
  TEARDOWN_CHECK receipt_exit=0 census_exit=0 clean_verdict=1 named_rules_absent=1 census_valid=1
  TEARDOWN_VERIFIED on all four conditions: zero exit AND successful census AND the product's own LAN_FIREWALL_CLEAN AND the owned named rules measured ABSENT by my census.
  stdout: LAN_BOOTSTRAP_NOT_UP + LAN_FIREWALL_CLEAN. The firewall mutation is REVERSED, established by measurement rather than by liam's report of it.

=== 5470 NEVER DISTURBED ===
GUARD_UNCHANGED at every bracket: post-build, a-1-post, a-2-post, a-3-post, pre-setup, post-setup, cleanup-pre-teardown, and final-post-cleanup. The elevated action is bracketed on BOTH sides, which is what F3 was added for.

=== CLEANUP: steps_failed=1, and it is the EXPECTED guarded refusal ===
cleanup-failures.txt exists and reads, verbatim:
  "2026-09-12T20:45:13Z isolated daemon stop refused or errored — REPORTED, not forced. See rig-daemon-stop.err"
rig-daemon-stop.err:
  "DAEMON_STOP_REFUSED: a broker-stopping command is never run by an spt endpoint ($OWL_SESSION_ID is set). this would stop the broker that hosts every spt-hosted endpoint on this node. No flag overrides this, --force included."
That is the guard working exactly as the procedure predicted. Pool RELEASED cleanly. Every other cleanup step ran.

=== THE ONE RESIDUAL, AND IT NEEDS A FRESH PID-SPECIFIC DECISION FROM YOU ===
The rig daemon is ALIVE and I have not touched it:
  pid 52608, spt.exe, CreationDate 2026-09-12T20:29:39Z -- inside this run's window, so the pid is authenticated by creation time and is not a reuse
  ExecutablePath C:\Users\decid\Documents\projects\spt-core\.worktrees\304-w2-repr\target\release\spt.exe
  It is the isolated-home daemon: fp-home/daemon.pid reads 52608, matching.
Note it holds the BUILT executable open -- a live demonstration of your correction that process-image absence never proved no file holds. Nothing on 29470 or 5470; it is idle, not serving.
I am NOT force-killing it. All prior elevated grants are closed and the 49768 grant is spent and non-transferable. If you want it gone I need a fresh authorization naming 52608 specifically, and I would want the creation-date guard re-checked at the moment of the kill rather than relying on this reading.

=== EVIDENCE ARCHIVED, PATHS UNCHANGED DURING THE RUN ===
105 evidence files copied AFTER END to .spt/preserved/304-w2-attempt2-20260912/, with the driver stdout alongside and a 106-line ARCHIVE.sha256 over the lot. The bundle stayed in place in the 59484d41 scratchpad throughout, as instructed. The scratchpad-reap risk I flagged earlier is now moot for this evidence.

No retries were run, no scope changed, no reconcile repeat, and no policy touched. Window closed at END; I hold nothing open but the 52608 question.