doyle → todlando. CODE GATE-PASS: w5_a2 Cut B + skip-to-live. Full source read, both crates.

w5_a2 Cut B: PASS. quiet_spawn_req faithful idle (sleep 600 / cmd /c "ping -n 600 ... >nul" — the >nul kills the ping-countdown-feeds-drain trap, correct). Assertion unchanged, teardown-only kill, verdict comment points idle-evict at REQ-HAZARD-DRIVEN-BY-IDLE-REMOTE-EVICT. Ruling executed exactly.

skip-to-live: PASS. Verified against the locked 3-part design:
1. Eviction signal DISTINCT from EOF: KIND_VIEWER_EVICTED, emitted in viewer_writer OFF the log lock. The flag-ordering is RACE-FREE — append does remove()→evicted.store(true,Release)→sink(tx) drops at scope end; store is sequenced-before tx-drop, channel-disconnect gives cross-thread HB to rx.recv()=Err, then load(Acquire). Writer always sees the flag before deciding. Mirrored correctly in resize_all. Socket-write-fail exits early (no marker, dead conn) — only rx-disconnect+evicted=true emits. Clean.
2. Resubscribe-from-floor: attach_skip_to_live = subscribe(from_seq=MAX, empty replay) → session_cursors.insert(sid,0) snap-above → next_seq=0. B2-SAFE confirmed: session_cursors is the serving brain's LOCAL dedup cursor, NOT delivered_through (advanced only broker-side by controller_writer.contiguous_advance); viewer intent = viewer row, no controller coupling. Units drive REAL read_event over real codec; cold_brain_still_rejects control proves the legacy reject-gap (controller B2 path) is untouched; snap-above test proves jump-accept + dedup-below + 3 caught mutations.
3. No-spin: serve_attach RESUBSCRIBE_INTERVAL=200ms real sleep-of-remainder, guarded to own session. HARD constraint satisfied.

traceable: REQ-VIEWER-SKIP-TO-LIVE-ON-EVICT [doc,impl,unit] activated, int correctly NOT-yet (rule 5, p0_paste on the combined fold). CONTEXT.md doc tagged. Clean.

seedmap hang: CLASSIFIED as the known Windows-full-suite-under-load contention class (kitsubito starvation + registry SQLITE_BUSY/busy-timeout-knob lineage) — passes in isolation, all your touched modules pass, seedmap untouched. NOT a b4/skip-to-live blocker. I confirm on the clean Linux matrix; if it repros pre-b4 it's separate CI-hygiene (file under that ledger).

NEXT: cut g2 inject-acceptance (dispatch_endpoint_input must not starve under output flood). What's your g2 ETA/approach? Once g2 is gated I run the COMBINED forkpty matrix (pump-carrier-fix + b4 + skip-to-live + g2): a_journaled + p0_paste + g2 + w5_a2, both runners ×3, then activate the int stages + fold to delivery-control → deployah v0.13.0. Commit+push is operator-gated (your call held for go) — I'll batch the fold so it's ONE matrix run unless the operator wants w5_a2+skip-to-live pushed now. Clean, precise work.