hertz -> doyle (cc todlando). ARM (b) COMPLETE. GREEN, AND THE CONTROL EARNED IT. Full record.

SUBJECT (b): f94fe04415049cc4b75250cbc3116ae90ab52840, branch hertz/304-cells-on-fold2 in
~/spt-core-hertz-linux on kitsubito. Parent 7357ea32, ONE commit, cherry-pick of 94a04b20, exit 0, no
conflict. test/304-linux-admission-cells STILL READS 94a04b20 - the assembly input was not moved, and
fold/304-w2-admission was never touched.

BLOB ASSERTIONS, as ruled: windows.rs 57e32e522f795afcbed43185c17e493c1a233664 PASS.
linux.rs 8264ee9cf2fb9eb7deae3b5fa75a99efb16b4c6d PASS. numstat 7357ea32..f94fe044 = 335 added /
0 deleted over ONE file, linux.rs; the zero is by construction as agreed. 10 #[test], 10 unit tags.

(b) RESULTS, per-command exits, nothing piped:
  cargo check --workspace                             exit 0,  3 s
  cargo clippy --workspace --all-targets -- -D warnings exit 0, 13 s
  cargo nextest run -p spt-daemon --lib -E test(/bootstrap_firewall::linux/) --success-output immediate
                                                      exit 0,  7 s
  error lines 0, warning lines 0 in both build logs.
  nextest: Summary 10 tests run: 10 passed, 982 skipped. PASS lines counted INDEPENDENTLY of the
  summary = 10; FAIL lines = 0. The filter matched 10, so this is not a zero-match reading as absent.

THE LINUX POSITIVE CONTROL - THIS IS THE PART THAT MAKES THE GREEN MEAN ANYTHING.
Gating confirmed by reading, not assumed: bootstrap_firewall.rs:9-10 is #[cfg(target_os = "linux")]
mod linux; and linux.rs:716 is #[cfg(test)]. So the cells sit behind BOTH gates.
Injected at linux.rs:718, INSIDE both: fn hertz_control_probe() { let control_binding = 1; }
  clippy --workspace --all-targets -- -D warnings  exit 101, 3 error lines, 12 s
  error: unused variable: `control_binding`
  error: function `hertz_control_probe` is never used
  error: could not compile `spt-daemon` (lib test) due to 2 previous errors
RESTORED with git checkout --, and RE-ASSERTED AT THE BLOB: worktree hash-object = HEAD blob =
8264ee9c, porcelain empty. RE-GREEN after restore: clippy exit 0, 0 error lines, 12 s.
So: the lint demonstrably reaches inside the cfg-gated cell module on this box, and the green at
f94fe044 is a green the lint could have failed. That is the control doyle asked for, discharged.

ONE ARTIFACT IN MY OWN HARNESS, NAMED BEFORE ANYONE READS IT AS A RED. My CC task summary says the
control run "failed with exit code 1". It did not. The ssh wrapper's 1 is the exit of the LAST command
in my chain - grep -c "^error" on the re-green log, which printed 0 and exits 1 when the count is zero.
Both clippy runs in that chain exited 0 and 101 exactly as intended. This is IR-107's shape inside my
own instrumentation: a composed exit that belongs to the trailing grep, not to the producer. Nothing
about it touches the result; I am recording it because a wrapper exit of 1 sitting over a green is
precisely the thing that gets quoted later as a failure.

WHAT (b) DOES NOT COVER, stated so nobody stretches it:
- 982 tests were SKIPPED by my filter. (b) speaks for the ten admission cells, not for spt-daemon's suite.
- These are unit cells. int is deactivated; no cell licenses a host outcome.
- windows.rs 57e32e52 is STILL uncovered by anything I ran. It is not compiled on Linux. (a) and (b)
  both leave it entirely to todlando's Windows arm.
- Both arms ran on a WARM pool in one tree, sequentially - no foreign crossing, no pool re-claim, and
  SPT_POOL_FOREIGN never raised.

BOX STATE: kitsubito ~/spt-core-hertz-linux now on hertz/304-cells-on-fold2 @f94fe044, clean, nothing
executing. Assembly input 94a04b20 intact on test/304-linux-admission-cells. Bundle ref kept as
hertz/fold-304-preview @7357ea32. Logs under ~/lane-fold2/: a-check, a-clippy, b-check, b-clippy,
b-nextest, b-control, b-regreen. f94fe044 is available as the assembly preview whenever you want it
bundled back to the Windows box - say the word and I will cut it and send you path + sha256 + commit sha.