todlando -> hertz, cc doyle. COLLISION DISCLOSURE, sent immediately because it bears on YOUR producer-3 numbers, not mine.

OUR MESSAGES CROSSED. doyle ruled at ~09:12:2xZ that I take the box now because you were still freezing and re-inventorying; you opened your producer-3 window at 09:12:45Z. I had already sampled admission at 09:12:57Z and launched probe (a). So a real-daemon cell of mine EXECUTED INSIDE YOUR WINDOW.

EXACTLY WHAT RAN AND FOR HOW LONG, so you can decide what it does to your attribution: one cell, spt-daemon::attach local_attach_via_loopback_conn_rides_the_same_pump, started ~09:12:57Z, 0.116 s of execution, 0.73 s of no-op build check before it. It spawns one in-process Broker, one findstr PTY child and a loopback conn under a private SPT_HOME and TMP in my scratchpad, on OS-assigned sockets. It is over; nothing of mine is executing now.
YOUR STATED CONDITION IS THEREFORE FALSIFIED, and you should not carry it as written: you recorded "Builder census ZERO, unreadable ZERO — the box is genuinely quiet this time, not merely authorized-to-overlap, so unlike the last window this producer's numbers are attributable without a caveat." For roughly one tenth of a second inside your window that was not true. Whether a 0.116 s single-broker cell can perturb a 300 s docs-server producer is yours to judge, not mine to wave away — but you cannot judge it if I do not tell you, and a caveat you discover afterwards is worth less than one you were handed.

I am NOT starting probe (b). It waits for your END, whatever you decide about producer 3.

PROBE (a) RESULT, since it is already in hand: GREEN. 1 test run, 1 passed, 15 skipped, 0.116 s, exit 0. Admission 09:12:57Z FREE=192,645,558,272; builders 0 and test binaries 0 at my sample, which is the same instant your window was opening — that is how the crossing happened.
Against the prediction file this eliminates outcome 1: the shared loopback pump is NOT broken by my daemon change, and the unmodified cell that references nothing I edited is the thing that says so. Remaining live outcomes are 2 (my helper) and 3 (product write-path on my change), and probe (b) with the N-1 control arm first is what separates them.