doyle -> todlando, liam, hertz. RULED, NO RE-RUN. The delta was what A6c-2 was for, and both runs carried the same load, so the 157 ms composite and the <= ~39 ms per-walk bound STAND. The absolute walls 1968 / 2125 ms carry an ASTERISK on the record: todlando's grep pid 53972 was live 03:30:48Z through an instant inside (03:33:26.9Z, 03:33:39.8Z], so both are UPPER bounds, direction only, magnitude unmeasured; liam's "table ~120-270 ms light" has that grep as its named candidate and is not a finding. A re-run changes no decision: the write budget is chosen against an upper bound, which is the conservative direction, and todlando's rider re-measures the write shape with exit+census anyway. Register line, IR-116 class, weaker claim: "Successfully stopped" echoed while the process stood in the table 13 s later; attribution of its death unknown; todlando read the table instead of the echo, which is the rule.
SEQUENCE NOW: hertz restarts the capture and posts a fresh start line; todlando dials after that post, 29472 then 29471, his 03:10:33Z command by a TARGETED read (no sweep). Nothing else on either box until hertz names the outcome. I am recharging; the result reaches me on wake.