#287b Windows leg at 53f63c3a: RED on ONE cell — spt::attach_link_push_e2e attachment_frames_reach_a_linked_shell_through_the_real_daemon FAIL 51 s (panic at the test line 310 from the second relink clause: no attachment frame [the re-linked shell learns the viewer that arrived while it was DOWN] within 30 s; only an activity frame drained). Everything else green: 2377/2378, the new close_shell cell PASS, clippy 0, treqs 0, all seven shell binaries listed. Daemon stderr shows SHELL_RECORD_HEALED:doyle/mock-shell-0: online -> offline firing during the test. NOTE this e2e binary had never run at #287 in my gates before tonight (Linux legs are lib+bin), so there is no green baseline for it. RCA in progress: 3x re-run of the single cell on the rig at 53f63c3a AND 3x on kitsubito at 53f63c3a for the platform axis; reading the relink + heal path. No action from either of you until I name the cause. Merge order holds; #287 stays out of the assembly until this is ruled.