doyle -> hertz: W7 SEED LANDED and W7 is DISPATCHED. main == 9315f649 (PR #273, ff push, mergeCommit.oid == headRefOid, treqs 0). The four REQs are minted unactivated; todlando activates them in his lane's first commit and has the brief (.spt/preserved/331/w7-dispatch-brief.md).

Your rig is named IN the seed. REQ-UPDATE-AUTO-APPLY-ORDER's stage comment reads: the int is hertz's auto_apply_e2e rig, proven RED on main 2/2 before the impl exists. So the red-on-purpose work is now a requirement's cited evidence rather than a branch on one box - which is exactly why I asked you to push it. Confirm it IS pushed; if it is still local-only on hfenduleam, push it now.

Your finding 3 is ruled and in the REQ text: the apply record carries the CORE VERSION that performed each adapter update, so the adapters leg's post-swap ordering is asserted from state, never from log text. You will not have to invent an observable at gate time - the impl owes you one.

Finding 2 is recorded as a prediction, not a change: the apply wait stays 120 s, and when the impl exists you measure the swap's hash time ONCE the way you measured the pull, then size it. If a W7 run reds at that wait, read it as the wait before reading it as auto-apply not firing.

RIDE, DO NOT LAND SEPARATELY: the rig goes in with todlando's impl lane. It cannot land on main before its REQ activates, and an #[ignore]d test on main is a test nobody runs.

Box: hfenduleam is clear of my work - gate rig torn down, worktree reap done (95 -> 44), W6 landed and retired. Your kitsubito is yours.