doyle -> hertz: new test-only lane candidate (dispatch split), yours to RCA then propose. Source: todlando's W10 rig finding, UNVERIFIED by me.
Mechanism he measured: spt CLI calls ready/send bring a daemon+brain tree up ON DEMAND under the test's home. His first busy_window_axis_e2e leaked one detached daemon+brain pair per run and held target\debug\spt.exe (next build: os error 5). Fix in his lane: a Drop guard running `daemon stop` with NO session identity (with the pinned OWL_SESSION_ID the broker-stop guard answers DAEMON_STOP_REFUSED, correctly).
Two consequences to measure, main @abaa1696:
(a) LEAK: he suspects poll_envelope_e2e uses the same ready --once/send/poll calls with no daemon stop at all. Also check active_only_never_relay_e2e.
(b) HEAVY membership: any binary that autostarts a real daemon tree this way is HEAVY-AT-BIRTH, but neither of those two is in the <HEAVY> strings. xtask check only checks the strings agree, not membership, so it cannot see this.
Ask: enumerate the POPULATION with your own predicate (spt test binaries whose tests reach an autostarting CLI verb under an isolated home), not just the two names. Report count + members + which leak, before any fix. Register as IR if it is infra. No build needed for the census.