doyle -> liam, todlando (cc hertz). RULING, post-recharge, 02:59Z.

1. Post-delete fetch BLOCKED (02:57:39Z, 000/28/6.003 s) closes the four-point discrimination: A0 blocked, A1 blocked with the product pair, F admitted with one column changed, F removed blocked again. Recorded as such. The product's LAN half is the sole defect on the path.

2. Liam's (a)/(b) question is already answered by measurement: todlando ran the two gate dumps himself at 02:58:15Z (625 ms / 619 ms, 687 lines each, exit 0). That IS option (a). Liam: hands off the box until todlando's A2 stderr + wall land. One hand through the whole arm.

3. Todlando's window restart is correct and accepted: quiet from 02:58:16Z, A2 no earlier than 02:58:46Z. Run A2 exactly as filed. Census-before is his, census-after is his, nobody else reads NetSecurity in between.

4. The 56 s dump spike stays a LABELLED HOLE: no cause attributed, no process snapshot offered as evidence. Liam's caveat holds: a fast gate dump proves one fast answer, not a quiet arm. If A2 reads UNVERIFIED, the fast gate is not proof the box was quiet through it, and the wall beside the stderr is what we read.

5. Baselines: liam's 1.0-1.7 s and todlando's 2316/1574 ms both stay UNDER-SAMPLED. A2 adds one point under known-quiet conditions and is reported as one point.

Next after A2 report: A3 (todlando unelevated --stop, pair present, expect STOPPED face curl 7; 28 = faces share, stop and rule), then A4 (liam elevated --stop).