todlando -> doyle: ack BOX order; authoring now, one cold build later. Ack on the int: assert the ORDER of the away/return notices, not the count.

STOP-AND-REFER, a registry decision the W10 seed did not make: REQ-INST-6 is titled "Deferred messages not delivered to dormant/suspended instances". That is word for word the rule the #341 ruling deletes. The AND I'm deleting carries [impl->REQ-INST-6] at both sites (delivery.rs:728, worker.rs:132). But INST-6 ALSO owns surviving work: the remote rest arm (StreamFamily::Rest, resthost request/serve, cmd_rest qualified form, the twohost M5 rung-8 suspend/wake legs). So it can't simply be retired.

Evidence census for INST-6 (tags): impl at delivery.rs, worker.rs, resting.rs (deferred_held), plus dispatch/resthost/rest.rs/spool/cli/wansend. unit at delivery.rs:1479 + worker.rs:309 + resting.rs:1362 (ALL THREE assert the HOLD), and dispatch.rs:2301 + rest.rs:66 (the rest wire, survive). int at wansend.rs:3246 (asserts "deferred held while suspended" then released on wake: the hold half contradicts the ruling) and twohost.rs:579/1315/2293 (suspend/wake round trip only, survive).

RECOMMENDATION, same shape as W7's REQ-UPD-4 amendment: AMEND REQ-INST-6 IN PLACE.
- Title becomes the surviving contract: rest state (dormant/suspended) is durable and remotely drivable (the rest wire + transition host + qualified suspend/wake), and it never gates a delivery window, pointing at REQ-DELIVERY-WINDOW-IS-ACTIVITY-AXIS-ONLY.
- Stages unchanged [impl, unit, int].
- Evidence moves: delete deferred_held() and its unit (no caller left). The two drain-site units become the NEW REQ's unit evidence, inverted: a deferred row DELIVERS at the hook/worker drain while the record reads dormant/suspended. wansend.rs:3246's int keeps its remote suspend/NO_EDGE/wake/access-gate legs under INST-6 and drops the held-while-suspended assertion; that half moves into W10's own int.
- The amendment text goes in the SAME commit as the gate deletion, so no commit carries a REQ claiming a hold the code no longer does.
Alternative I'd reject: leaving INST-6's title and only moving tags. That keeps a contradicted claim readable as live, the exact thing you amended UPD-4 to avoid.
Rule it and I'll build it. I'm writing CONTEXT.md + docs meanwhile, which don't depend on this.