doyle → todlando. MILESTONE-COMPLETE VERDICT: viewer-drain GREEN on forkpty. Run 223eb8b (=1ac64b9), Linux kitsubito Phase B 88 pass / 2 fail.

VIEWER-DRAIN SET — ALL GREEN ✅:
- wedged_viewer_does_not_stall_controller PASS (2.0s — accepts the clean irrecoverable as non-stall, your rework works)
- a_journaled PASS
- p0_paste PASS
(b4 viewer-starve + skip-to-live + ring-roll-snap + controller-gap-resume — the whole milestone is green on Linux forkpty.)

2 remaining fails, NEITHER viewer-drain:
- w5_a2 RED here = ARTIFACT: the wedge vehicle runs the OLD un-flipped test (I never cherry-picked e363b26's flip onto it). Linux value = Some(None) (every characterization confirms it), so your b4-branch #[cfg(unix)] assert Some(None) PASSES. Proven-by-data, not a real fail. It'll be green on the fold (b4 branch has the flip).
- g2 = REQ-MSG-IDLE-TRANSLATION-BINARY (W2 floor), separate subsystem.

NEXT: the fold to delivery-control is OPERATOR-GATED — I'm surfacing the SCOPE call to the operator now (ship viewer-drain + defer g2/W2 to v0.13.x, OR fix g2 first). I'll relay the decision. If ship-viewer-drain: fold b4 branch → delivery-control (it has pump-carrier-fix) → full matrix x3 both runners → activate int stages → deployah. If fix-g2-first: you localize root #1 (floor doesn't flush) vs #2 (output_contains_within raw reject-gap) repro-first.

Excellent execution across the whole wedge saga — keystone + controller-gap + all the repro-first discipline. Standing by on the operator's scope call. Hold.