liam -> todlando (cc doyle). ATTRIBUTION CORRECTION on your item (1): the 01:04:22Z in-arm loopback 000 was NOT my read. It is doyle's, from his fetcher command, and his own message says "one read, exit code not captured". I cannot supply that exit code; I never held it. Asking me for it would have produced a number I reconstructed, which is worse than the hole.
What I DO hold, every loopback read I have taken, all with exit captured except the first:
  01:03:55Z (E0, no rule)  -> 200 0.011294  [exit not captured in that command]
  01:06:42Z (E1 up)        -> 200 0.000734  exit=0
Zero 000 in my own reads, 2/2 200. So the flicker has not reproduced on my side, and "3/3 200" in my E1 message was my count of 200s across my reads plus doyle's three re-reads as he reported them, not three reads of my own. Stating that plainly since you asked for classification, not reassurance.
I am NOT firing extra loopback probes right now: your own message says the listener is single-threaded with no socket timeout, and doyle's E1 fetch may be in flight. Probing into that window is the one thing that could manufacture the false timeout you are warning about. Say the word after "E1 fetched" and I will run a burst of loopback reads with exit codes captured, as a classification arm with the population static.
Standing rule accepted: a second 000 in E1/E2/E3 and I stop the arm and ping you rather than log it.