todlando -> doyle. WINDOWS ARM COMPLETE AT 7357ea32. GREEN, AND THE CONTROL EARNED IT.

SUBJECT: 7357ea327c79f5930bdfe5947d427eabd19cf9b8, .worktrees/304-fold-admission, pool claimed for this
lane at that base, free space 157G. Verified at the arm, not carried forward:
  windows.rs blob 57e32e522f795afcbed43185c17e493c1a233664
  git status --porcelain = EMPTY

RESULTS, exit per command. Where a pipe appears the exit is PIPESTATUS[0] -- the PRODUCER's own, never a
composed trailing exit -- and the last two arms were run fully UNPIPED to a file as you asked:
  cargo check  --workspace --all-targets                exit 0
  cargo clippy --workspace --all-targets -- -D warnings exit 0, 1m10s
  cargo nextest list -p spt-daemon -E test(bootstrap_firewall)   exit 0  -> POPULATION 14
  cargo nextest run  (same filter) --success-output immediate    exit 0, 7s
  cargo check -p spt-daemon  (RE-GREEN after restore, unpiped)   exit 0, 7s, 0 error / 0 warning lines
  cargo nextest run  (RE-RUN, unpiped to file)                   exit 0, 210s incl. rebuild

POPULATION BEFORE RUN, as you required: list enumerated 14 named tests; the run reports 14. Not a
zero-match filter. Independent of the summary line, PASS lines = 14, FAIL lines = 0, from the UNPIPED log.
  Summary: 14 tests run: 14 passed, 1232 skipped.
ONE THING I AM CORRECTING BEFORE YOU READ IT ANYWHERE: my first pass reported "PASS lines: 6" against a
summary of 14. That was NOT a discrepancy -- it was my own `tail -60` truncating the earlier PASS lines.
An independent count taken from a truncated log is not independent, it is just wrong. That is why I re-ran
unpiped; 14/14/0 above is the real count. Do not carry the 6.
The 14 include the two cells that assert the constant FOLD-2 documents:
  a_program_bearing_rule_does_not_satisfy_a_spec_that_wants_no_program_filter
  the_write_path_renders_both_halves_and_no_program_filter_by_default

THE WINDOWS POSITIVE CONTROL, DISCHARGED. This is the part that makes the green mean anything: hertz's (a)
and (b) are both green with his own Linux control, but windows.rs is cfg-gated and NOT COMPILED ON LINUX,
so blob 57e32e52 is covered by NOTHING except this arm.
  pre-state: blob 57e32e52, porcelain empty (the restore is only safe because FOLD-2 is committed)
  injected `const SPT_CONTROL_MUST_FAIL: u32 = "not a u32";` beside DESIRED_PROGRAM
  cargo check -p spt-daemon -> exit 101
    error[E0308]: mismatched types
       --> crates\spt-daemon\src\bootstrap_firewall\windows.rs:225:36
    error: could not compile `spt-daemon` (lib) due to 1 previous error
  It went RED and it NAMED windows.rs. The compiler demonstrably reads that file on this box.
  git checkout -- restore -> blob back to 57e32e52, 0 modified lines, porcelain empty
  re-green -> cargo check -p spt-daemon exit 0, 0 errors, 0 warnings
So the green at 7357ea32 is a green that could have failed.

IR-116 WRITTEN, as you asked. Branch docs/ir116-guard-false-green off current main 9a39e8da, ONE commit
15460d706f952c0200d140621b09bf1671dc6906, one file (docs/INFRA-REGISTER.md, +10), house bullet format,
trailer in the space spelling. Deliberately NOT on the fold lane -- you verified that as four files and I
am not making it five. Unpushed; say where you want it landed.
It records the guard mechanism, the TEST_EXIT=137-is-void corollary, and a second-order slip I made INSIDE
the entry: `git checkout -- <file>` restores from the INDEX, not HEAD, so "reverting" a staged first draft
and appending a rewrite produced a file carrying BOTH. The diffstat caught it -- 49 insertions for a
ten-line entry. Fixed with `git checkout HEAD --`; the committed diff is 10 insertions, IR-116 appears once.

STATE: fold worktree clean at 7357ea32, pool still claimed for this lane (not released -- tell me if you
want it dropped). Nothing of mine executing. hertz's f94fe044 is available as the assembly preview.