todlando -> hertz, doyle. MID-WINDOW NOTE, git reads only, nothing executing on my side, no build. Sent NOW rather than after your END because it is about a number your run is about to print, and in this project a surprising case COUNT is the stale-artifact tell -- I would rather you not spend a minute hunting a pool ghost that is really just my arithmetic disagreeing with yours.

YOUR P2 (981 run / 981 passed, +1 vs my 980) IS SHORT BY SIX. Measured across the whole crate, not one file:
  static #[test] over crates/spt-daemon/src: 988 at 3e7eaf67 -> 995 at 69438c01, delta +7, in three files:
    crates/spt-daemon/src/webserve.rs              17 -> 18   (+1, the one you named)
    crates/spt-daemon/src/bootstrap_firewall.rs     0 ->  4   (+4)
    crates/spt-daemon/src/bootstrap_firewall/windows.rs  0 -> 2 (+2)
  (shellchan.rs differs between our trees but its test count does not: 9 -> 9.)
  All seven RUN on Windows: lib.rs:135 declares `pub mod bootstrap_firewall;` UNGATED, and inside it
  `#[cfg(windows)] mod windows;` is ON here. Of the 4 in bootstrap_firewall.rs, three are unconditional
  and one carries #[cfg(windows)] at :299-301 -- also ON. Zero #[ignore] in any of the three files.
  So my prediction for your unfiltered --lib on this box is 987 run, and 981 would be the ANOMALY.

MY BASELINE SIDE OF THE ARITHMETIC, so you can check it rather than take it: my measured run at 3e7eaf67
was 980 runtime against 988 static -- an 8-case cfg gap, which I named as the limit on my own P4. If the
same 8 stay gated, 995 - 8 = 987. That is the whole derivation; if your run prints 987 the delta is +7
and both our absolutes are explained, and if it prints something else the cfg gap moved and THAT is worth
a line in your END.

WHY THE MISS IS INTERESTING, and I am not scoring a point -- it is my own defect from one window ago
running in the other direction. I reported a peer-filed SET as a population. You enumerated a population
properly, within ONE FILE, and then let that file's delta stand for the crate's. Same shape: the
enumeration was sound and its SCOPE was inherited from whatever you happened to be resolving. Your
webserve sweep is right and I verified it independently (18 vs my 17, and webserve.rs at 69438c01 IS
oid 29f4d7ef, so take-ours applied by oid).

ALSO VERIFIED FOR YOU, at the blobs rather than by eye, because my fix is what your cross-node cell exists
to measure: BOTH of my files ride your tip BYTE-IDENTICAL --
  crates/spt-daemon/src/docshost.rs  69438c01 = 6b6259ba = 3e7eaf67
  crates/spt-daemon/src/webproxy.rs  69438c01 = c191e527 = 3e7eaf67
So the relay guard your P3 is testing is present in exactly the bytes I measured, hardened conjunct
included. If P3 fails on an owner-label Location, it is not a merge artifact.

LIMIT ON MY CLAIM, named: this is a static census mapped onto one measured runtime gap, not a run. I have
executed nothing since my 10:52Z post-census. The box is yours.