todlando -> doyle: W10 HANDBACK. fix/341-busy-window-axis @ 89fd44127e0cc77a00a54edd9fab9481f4892a43 (pushed; ls-remote confirms). Off 9a5dd7be: 7d62d3d8 activation, 89fd4412 the change.

EVIDENCE (hfenduleam, box yours->mine per your GO; census at 02:42:37Z: no cargo/rustc/nextest/link/Runner.Worker, free 131,379,220,480 B):
- cargo check -p spt-daemon -p spt --tests: exit 0.
- xtask gen: reference.md + changelog.md regenerated. The reference diff is only --active-only -> --busy-only; neither alias rendered. xtask check: OK, exit 0 (captured to file, exit read on its own line).
- clippy same scope -D warnings: exit 0, 0 warnings. treqs: exit 0, 0 must; still 5 unjudged files (same count as before; set NOT verified).
- Units: spt bin 49/49, spt-daemon lib 18/18. By name: poll_drain_delivers_the_busy_window_in_every_rest_state, worker_poll_delivers_the_busy_window_while_resting, busy_only_is_the_spelling_and_the_old_ones_parse_hidden, remote_rest_suspends_and_wakes_over_the_wire (INST-6 int, wire legs only).
- INT GREEN: busy_window_axis_e2e PASS (last clean run 7.06s). active_only_never_relay_e2e PASS (moved text; still sends --active-only, so the alias is exercised end to end).
- NEGATIVE CONTROL (the AND re-added verbatim in effect at poll_drain), read from logs:
  int: 03:03:26Z-03:03:42Z, FAIL exit 100, panicked at busy_window_axis_e2e.rs:170:5 = the [int->] tagged assertion "the away notice and the peer message drain WHILE dormant (releases#341): []", left 0 right 2.
  unit: FAIL at delivery.rs:1507, "dormant drains the busy window on its own poll, oldest first", left [] right [away notice, peer while unwatched].
  Restored with sha256 identity both times.
  Note: the ORDER is carried by the arm structure. The away notice must drain on a poll taken BEFORE deliver_away_notice(Back) runs. The control reds that poll, so pre-fix the away notice cannot arrive while away. That is #341(a) exactly.

A FINDING FROM MY OWN RIG, fixed in the lane: my first e2e version LEAKED a detached daemon+brain pair per run, holding target\debug\spt.exe (the next build failed os error 5: that was my first control attempt's EXIT 101, which never ran). The test's CLI calls start a daemon on demand. My StopDaemonOnDrop guard first ran `daemon stop` with the test's pinned OWL_SESSION_ID, and the broker-stop guard correctly answered DAEMON_STOP_REFUSED. The guard now stops with no session identity. Censused after green, relay, and control runs: 0 processes from my pool each time (7 leaked PIDs killed along the way, all from my pool by exe path). UNVERIFIED suspicion, not mine to fix: poll_envelope_e2e uses the same ready --once/send/poll calls with no daemon stop at all.

REGISTRY: INST-6 amended per your ruling (title points at the WINDOW REQ, no negative restated; D5a comment rewritten, D5b kept; DEFERRED.md tag untouched). CONTEXT.md gate clause replaced positive; busy-window sweep; dormant's two arms. Docs: messaging, instances, lifecycle. CHANGELOG Fixed + Changed.

NOT RUN: Linux / kitsubito. A live away notice from a real 5-minute detach (the int drives the real producer fn in-process, not its timer). The fleet's other adapters' use of --active-only (aliases keep parsing; not exercised beyond the relay e2e).
Pool .worktrees/w10-busy/target still claimed. Release + IR-150 teardown after your gate.