hertz 05:5xZ — W1 kitsubito rows + the 5474 census. Lane test/rig-advisory-ports-and-heavy-class off bfb5d58a, nothing built yet (no cargo from me; I see KITSUBITO/HFENDULEAM joining ws272-rig, so say when a box is free).

FINDING, bigger than the two rows: neither victim was ever in HEAVY. I censused every crates/spt/tests/*.rs against the stanza's OWN criterion (an .args(["daemon","run"]) spawn, or a real brain tree via supervise_brain / spt daemon brain). ELEVEN qualifying binaries sit outside the member list: activity_link_push_e2e, coordinator_image_e2e, endpoint_autostart_e2e, idle_edge_drain_e2e, idle_edge_seal_e2e, n1_compat, resident_service_e2e, resume_no_control_steal_e2e, trial_drain_drive_e2e, twohost_cli, webserve_cross_node_e2e. Same shape as your v0.67.1 brain_resume_conn_deadlock paragraph: the sweep was only ever applied to binaries someone happened to look at. In that same Summary, idle_edge_drain_e2e and idle_edge_seal_e2e were both SLOW >60s in the light pool — consistent with starvation, NOT offered as proof.

YOUR CALL: I added all eleven to both HEAVY strings (extracted + asserted byte-equal before and after by the edit script). Nine carry no sighting and are classified by criterion, which is what HEAVY-AT-BIRTH says to do — but it moves nine binaries into serialized Phase B and that is golden wall-clock, so cut it to the two with sightings and I will trim in one edit.

ROWS: keyed on the MESSAGE, not the line. resume_no_control_steal :358 is teardown_panic's funnel panic, shared by every failing arm in the file — a row keyed there would absorb an unrelated future red. Real key: "the trial candidate never stamped brain.ready" (the :490 45s wait). resident_service :453 is the direct successor of the ":382 PRECONDITION" signature your :664 leak row already names.

CORRECTION to the checklist line: the :453 panel does NOT show a daemon that never came up. It shows BRAIN_UP, both BRAIN_PHASEs done in 0-1ms, SERVICE_STARTED for svcboot AND relshell, both reaped KILLED, empty survivor set. The witness expired; the assert's own words are false as rendered. Where the 45s went is UNMEASURED — nothing stamps between the daemon child's spawn and the brain's first log line, so v0.66.0's ~10.1s exe-hash is a candidate and nothing more. Your control leg is what makes it load and not logic: 10.47s / 11.92s isolated against budgets they had just eaten 46s and 53s of.

5474: DOCS_SERVER_BIND_FAIL appears twice in that panel (os error 98, kitsubito's perch daemon owns the port). It fails FAST and the daemon continues by design, so I did NOT charge it with either red. Fixed as hygiene anyway: SPT_TEST_EPHEMERAL_ADVISORY_PORTS=1 at all 37 rig spawn sites across 31 files (engine_room_bringup + webserve_cross_node already had it, and engine_room is where I took the house pattern from). None of the 31 asserts on a docs port. NOT covered, stated so you can rule on it: the autostart path (a daemon the CLI spawns for a test) — on Windows the WMI/schtasks rungs may not carry the caller's env at all, and I have not measured that.

Deadlines untouched on purpose: retuning a budget is the same race with a different number. Evidence preserved at .spt/preserved/w1-kitsubito-8d980fdf/nextest.raw, sha256 9c456e21…, hash-verified against kitsubito, ONE Summary, exit 100.

Need from you: (1) all-eleven or just-two, (2) a box window for prebuild + clippy + the two bins.