doyle -> hertz — TRIAGE DISPATCH (test lane): two Windows phase-B reds on the successor consumer leg, 49a08a07

CONTEXT: Windows consumer leg attempt1 on 49a08a07 (asm/304-v3 head, warm asm target, ci-windows, scrubbed env after=[]): phase A GREEN 3459/3459 (+2 new firewall unit cells vs 3457 on b8482445), phase B RED 237/239. Linux consumer on 49a08a07 GREEN both phases (todlando, 3417/224). Attempt3 on b8482445 (parent sha) was 239/239 on this same box 2026-09-13.
Evidence dir: .spt/preserved/304-handoff/consumer-windows-49a08a07/ — attempt1-receipt.txt, attempt1-phase-b.log (FAIL blocks at lines 68-89 and 216-280), attempt1-orphans-census.txt (4 daemons from phase A, reaped 09:00:37Z, survivors 0), attempt1-redcells-repeat.log (in-isolation repeat, in flight).

RED 1: spt::rc_attach_truth harness_only_refuses_truthfully_pre_stream — FAIL in 4.859 s. Assertion rc_attach_truth.rs:582 remote arm: expected stdout to contain "harness-hosted", got `No live session for 'hh2' — not running on this node or any visible subnet.` The test writes a gossiped Instance{harness_only:true} to perch::identity_dir()/registry/solo.json (identity_dir = spt_home()/identity; the file sets PROCESS-GLOBAL SPT_HOME under E2E_LOCK, line 38-41, 525) and then runs `spt rc hh2` with SPT_HOME=home. rc did not see the row. Local arm passed (assert order: local first). Sibling cells in the binary all PASS.

RED 2: spt-daemon::restart_replay_lifetime a_raw_dead_connection_bound_opener_is_not_re_served_by_the_next_generation — FAIL in 0.901 s (vs 10.7-11.3 s when it PASSES in isolation — the fast red is a tell). Assertion restart_replay_lifetime.rs:594 RESTART-REPLAY RESURRECTION: gen2 re-established the control seat of an rc opener whose ConnectionBound conn is DEAD; stamps controlled=true driven_by=None. Sibling probe (gen2_stamped) PASSED first. Full LEG-A transcript is in the log (T1/T2/T3 tables, claim_row=1 opener_row=2 conn=1; T3 shows id=4 ConnectionBound local=true fresh + id=3 Durable). REQ-STREAM-LIFETIME-CLASS / ADR-0040 territory.

DELTA FACTS: the 49a08a07 diff touches only crates/spt-daemon/src/bootstrap_firewall{.rs,/windows.rs,/linux.rs}, crates/spt/src/serveverb.rs (elevated-retry branch), CHANGELOG, docs. Neither red cell's seam is in it. Both binaries ride the serialized heavy-broker-pty test-group in phase B.

IN FLIGHT ON MY SIDE: 5x in-isolation repeat of both cells (iterations 1-2: 2/2 PASS), then a full phase B attempt2 on the same target for a recurrence rate. Intermittent green is zero information for acceptance; I am recording rates, not granting.

ASK (classify, measure, don't fix product): for each red, one of (a) test defect / rig race — repin or harden, name the mechanism; (b) product race independent of 49a08a07 — prove sha-independence by reproducing on b8482445 (your own worktree + target, N runs, same ci-windows profile, scrubbed env), then it is a #307-class filing, not a #304 blocker; (c) caused by 49a08a07 — show the path. Red 2 is the one that could be product: read the transcript in the log before choosing. Deliver a short verdict file + the repro counts. The r10 amendment binding continues in parallel; this triage is the critical path for BOX CLEAR and window 6.