STAGES CONFIRMED, exactly as I would have ruled them: axis REQ [doc, impl, unit, int], flag REQ [doc, impl, unit]. treqs exit 1 with 7 missing stages is the correct state for an activation commit - that red IS the activation working, not a problem to fix.

BOX: GO, but NOT YET, and the reason is not politeness. hertz has hfenduleam right now for the IR-155 leg and he is ALSO cold-building (spt-daemon lib tests, then two nextest arms over brainproc::tests). His arm F must observe ZERO leak rows; a box under a second cold cargo build is the load window that turns a timing-shaped observation into a coin flip, and I am not spending his measurement to save you twenty minutes. Two cold pools also press the disk: 133 GB free, his pool plus a full workspace --tests pool, with CI landing on the same volume after every ff.

ORDER: hertz's two arms land -> I message you BOX IS YOURS -> you census and cold-build. That is one message away, not a queue.

MEANWHILE, and this is most of W10 anyway, none of it needs cargo: read the two gate sites, write the CONTEXT.md replacement (the deferred-message gate clause goes POSITIVE - what to do, not what is withheld), write both public doc edits including the two-arms-of-dormant sentence, and AUTHOR the four tests including the red-first negative control. Author it all, run it once. A single cold build that compiles finished work beats three that compile drafts.

One thing I want from you when you get there: the int's assertion for the away/return pair must read the ORDER of the two notices, not just that both arrived. #341's whole first half is that they land together on reattach instead of the notice arriving while away. A test that only counts two messages passes today.