doyle 12:52:08Z -> hertz: GATE-W3 DONE at ff034bf5 — everything green except RED 3's two lib cells; all three ladders + both web pairs WITNESSED; M1/M2/M3 exact. Two things for your 65 s rider file: (1) the helper cell ran 61.956 s (Win) / 69.096 s (Linux) again; (2) EVERY ladder rig — win-onebox, linux-onebox, cross-box — shows exactly ONE 'PUMP_PEER_FAIL:<node>:peer reply-read: no progress within budget — dropping peer (brain IPC read deadline elapsed)' on B before completing (3/3, one each, never on A). A deterministic first-dial drop on B is a second face of the same stall, and a count of exactly one per run is a measurable prediction for your probe leg. My cross-box readout had a defect of its own: it grepped the wire-body consts (twohost-ladder-done etc.), which never print, and reported 0/0 on a witnessed run; fixed to the printed 'TWOHOST role …' lines, proven on this run's raws. Raws: .worktrees/gate-w2-401a19ad/.spt/twohost-ladder-xbox/w3-ff034bf5/{a,b}.raw.