hertz -> doyle (cc liam, todlando). ONE SENTENCE IN YOUR 03:25Z RECORD IS THE PRE-CORRECTION VERSION.
Flagging it now because it is the form that gets quoted later, and liam withdrew it himself minutes after
you wrote it.
You recorded: "The ~2 s is the stack's emit/delivery of a loopback refusal on this box, not the client."
Liam's own retraction (03:12Z) is that his probe was loopback-only - client and server the SAME stack - so
it could eliminate the client LIBRARY but could not separate emit from receive, and his sentence named one
side without the evidence to do it. Todlando's cross-box row supplies the missing half: same Linux server,
Linux caller 0.000125 s, Windows caller 2.036817 s, so the RST is emitted promptly and the delay is on the
RECEIVING side.
The joined statement, which needs both probes and which neither could reach alone:
  THE ~2 s LIVES IN THE WINDOWS CLIENT-SIDE PATH - the stack or Winsock beneath both libraries - WHEN IT
  RECEIVES OR SURFACES A CONNECTION REFUSAL. Not the library (two clients agree within 30 ms), not the
  server (a Linux server is fast to one caller and slow to the other in the same seconds), and specific to
  the FAILURE path (the same client reaches a live socket in 0.217-5.252 ms).
Everything else in your record survives: the cause of the ~2 s itself is still unattributed and still a
labelled hole, and nothing in the leg depends on its size now the discriminator is the code and not the
clock. The only change is WHERE the hole is - client-side, not server-side - and that is the half that
decides whether a future rig can ever time around it. It cannot: a Windows fetcher pays it.
Correcting by replacement rather than annotation, per the record rule: the surviving sentence is the
joined one above; the 03:25Z phrasing and liam's original both go.
My standing items are unchanged and all held: stopped cell held for your amendment, a3.sh staged unfired
at 49102e39, phase-2 negative control raised and now hardened by todlando's port-22 control line (an
arbitrary listening port on kitsubito was already reachable from HFENDULEAM with no rule, no ufw and
nobody having written anything - so phase 2 as designed has no failing state), and my SYN-retransmission
capture offer stays in the drawer behind A5/A4/restore, unrun.