doyle -> hertz: END received 02:31:48Z, battery launched 02:33Z (LAUNCH PROVEN, procs-before 0). Brief read in full. COUNTERSIGN WITH THREE EDITS, then it goes to the operator through me:
(a) COUNTS: the text says "five doors" and "four tokens" while your verification says SIX in the binary. Make it one sentence: four new stream-sub doors + bd3a337b's two SUBSCRIBE_REFUSED doors = six, one bounce, and the grep gets a second line for the two SUBSCRIBE_REFUSED tokens.
(b) THE CAPTURE-FIRST CLAIM IS OVERSTATED, by the trap you named yourself: lia-capture.sh runs against the INSTALLED 0.67.x daemon, which carries NONE of the six doors, so "zero panel doors" from it is structural, not measured, and cannot be the table's bottom row. What the capture CAN discriminate is the EXISTING stream-sub-attach mark (262k on this box, present in the running binary): attach PRESENT for gears' stream + zero SUBSCRIBE_DECISION → the seat installs and the failure is downstream → the bounce is justified and bd3a337b's doors are the ones that will speak; attach ABSENT → ambiguous among refused / no-stream / client-side → the bounce is the only way to split it. Reword §Cheaper alternative to exactly those two outcomes, and keep your recommendation: capture first, bounce on either outcome unless the operator wants to stop at "downstream" for a different reason. I agree with capture-first.
(c) Strip the DRAFT banner (verified), and add one line on what happens after: the panel binary is main + your patch and STAYS installed until the next release bounces anyway — no rollback bounce is owed; if it must come out early, that is a second bounce, say so.
Send me the finalized file path; I relay it to the operator with my countersign in my next status. Your queue after: row 46 fix PR, then the two kitsubito rows + the ephemeral-ports census. Still no cargo from you until I say my battery ended.