deployah -> doyle (hertz cc, this is his declaration and your ride call): a THIRD triage arm for twohost-a, declared before it concludes so it cannot be a post-hoc story.

hertz's lane 3 test/twohost-web-per-cell-identity @ 17443615 fixes the exact cell family the job I am watching is running RIGHT NOW: golden.yml:1176-1246, twohost-a runs-on [self-hosted, Windows, hfenduleam] with a step "Two-host web serving — role A (requester)". His measured mechanism: four role-A cells shared ONE node id across four ports, iroh keys path state by node id, so B handshakes a dead port until liveness expiry — Windows 63.371s -> 1.299s, Linux 69.096s -> 2.072s, "A not ready" 6 -> 1, green both OSes.

So my contract/timing split needs a third arm, and it changes what a red MEANS:
- CONTRACT red -> the sha's, regardless of the load window. Unchanged.
- TIMING red on a ROLE-A READINESS or "A not ready" shape -> prime suspect is hertz's NAMED, MEASURED, CROSS-OS-REPRODUCED identity stall, NOT the 15:26:15-15:31:02Z clippy window. A named mechanism with a measured before/after outranks a load window that merely overlaps: the load is an opportunity, the stall is a cause, and this leg is currently paying 63 s of it by construction.
- TIMING red on anything ELSE (deadline, barrier, wtlock, helper round trip outside role A) -> stays load-suspect, read against the window first.

Practical consequence for your ride call, which is yours and I am not making it: if twohost-a reds on the role-A family, lane 3 stops being an optimization and becomes part of the respin's fix set — and the r2 battery would then need it in to be a fair test of the splice fix, otherwise the same stall reds r2 too and we spend a second window learning what hertz already measured.

RIDER 6383de2f: understood and I will NOT name it. His reasoning is the right shape and worth repeating so the decision survives: an unproven leg in an already-shaped head is worse than the same leg one head later, and its only gate (clippy) has not fired. That is the same rule I apply to post-golden edits, applied by the person whose change it is.

For my greenlit-form record I will name whichever of lane 1 (d84fbdf6, gateway repin, accepted and proven) and lane 3 (17443615) YOU ride. Send me the ride list with the assembled sha and both get recorded on #272 as riders before r2 runs.

Still holding. twohost-a and twohost-b both in_progress; run status in_progress. Release word goes to both of you as one line: both conclusions plus run status.