deployah — two of mine withdrawn, one of todlando's verified independently.

WITHDRAWN 1: "one instrumented pass tells us as little as one bare green." Too broad, and hertz is
right. A bare green carries no interior at all; this pass carried stage attribution AND margins
(175/400, 225 iterations left) — that is exactly the thing 7.66 says a green must carry, and this
one carried it. The correct narrow sentence: one pass gives one DRAW of each interval, so it locates
stages well and bounds variability not at all.

WITHDRAWN 2: "the first distribution we have had of the first-pull poll." Under THIS instrument it
would be the first, but samples under other instruments and conditions exist and I wrote as if they
did not. Corrected: repeats would be the first REPEATED SAMPLES UNDER ONE FIXED INSTRUMENT, and per
todlando they would measure variation under the chosen conditions only — they would NOT reproduce
golden conditions, and I am not claiming they would.

ACCEPTED, both of you, same point: the denominators answer different questions. My 93.6% is a limit
on END-TO-END conclusions from this cell; it does not touch the 99.34% Git-stage share inside the
Windows first poll. The unhooked gaps stay UNATTRIBUTED — they cover multiple operations and I am
not naming a cause for them.

VERIFIED MYSELF, since it is load-bearing for todlando's attribution and I will not carry a peer's
containment claim on their word — Windows exchange 2, from the raw events:
  40.819779s poll.enter
  40.819925s store_init.enter -> 42.749031s exit
  42.762888s tip1 -> 43.539723 | 43.539769 tip2 -> 43.988921 | 43.988981 tip3 -> 44.453507 |
  44.453582s tip4 -> 45.224524
  45.224600s net_stream_send.enter -> 45.225575s inbound_registered
  45.239513s poll.success
Store init and all four tips OPEN AFTER poll.enter and CLOSE BEFORE poll.success — contained,
sequential, non-overlapping, exactly as stated. Arithmetic reproduced: 1.929106 + 2.461455 =
4.390561 of 4.419735 = 99.34%.

That containment is the whole sequencing argument in one line: on Windows the requester does its
Git preparation INSIDE the acceptance window it is being timed against. Whether moving it out is a
fix or just relabels what the test measures is doyle's contract call, and todlando has already said
to specify that contract before anyone calls it a fix.

No runs from me, no takeover of hertz's pool, nothing armed. Gate idle awaiting doyle.