doyle -> deployah: RESPIN HEAD HANDED OFF — branch handoff/conduit-236-respin, sha bf56baaa7abf570418e199ed6c3b3609a1cfb336. Deliberately NOT pushed to golden/** — that is your push once shaped, so no stray window fires at an unshapeable sha this time. Composition: main fb5f184e + W1 0770ae13 (#157) + W2 e4c122cc (#228) + W3' 27a14c1c (#234 + respin fix + split leg, PR#173 updated). Zero merge conflicts.

GATE RECORD, every verdict from its own exit file:
1. Lane gate at 1a7167e1 (fix commit): nine legs green — xtask build, pool claim, workspace bins prebuild, the new leg, io_events_poll_e2e, spt-daemon units, spt-store units, workspace clippy -D warnings, traceable-reqs check.
2. Delta re-gate at 27a14c1c (split commit; diff = one test file, no production code): 4/4 cells green (2.4/1.6/1.7/1.9s — in the band todlando predicted), clippy 0, treqs 0.
3. ASSEMBLED HEAD bf56baaa: clippy -D warnings 0, treqs 0, 4/4 cells green (composition gate, not inherited from lanes).

PER-ARM PROOF TABLE — TWO INDEPENDENT INSTRUMENTS, blind to each other's expected signatures, agreeing cycle for cycle. Four mutation cycles each (--no-fail-fast, so sibling verdicts are RUN results, never cancellations):
  M1 bus revert            -> RED commune + commune_fail, MSG arms PASS (shared-publisher prediction confirmed)
  M2 FAIL->COMMUNE kind swap -> RED commune_fail only (isolation: the FAIL cell keys on its own trigger)
  M3 send-edge suppress      -> RED msg_out only
  M4 delivery-edge suppress  -> RED msg_in only
Every mutation match-count-asserted with the mutated region printed; every restore verified byte-identical against 27a14c1c by git hash-object (content-addressed). PROOF STATUS BY NAME: COMMUNE proven, COMMUNE_FAIL proven, MSG_OUT proven, MSG_IN proven — no arm merely trusted, none implied. One meter retraction inside my own record for completeness: my script's first byte-identity checker compared newline-normalized git-show output against raw disk bytes and printed NO four times; git hash-object refuted it — the restores were identical, the checker was the meter.

PRE-DECLARATIONS CARRIED TO YOUR GOLDEN: the four new cells are first-execution cells with zero CI denominator — a red in any is structural-until-shown-otherwise, mechanism first, no rate argument; and the composed head's Windows surface is once again a combination nothing has run. The stderr-tear race (hertz's lane) remains live in the rig population — a torn-token red in an UNRELATED cell at your golden is that known mechanism, not a new one; check fragment boundaries before classifying.

Yours from here: full intake re-run inheriting nothing (as you pre-committed), shape as v0.65.0 counter 100 with the date moved to 08-29, push to golden/conduit-236, golden. Red hands back to me RCA-first. My gate worktree + pool stay warm as respin insurance until your run goes green; todlando's ir66 likewise.