**FOLD-2 gate at `7357ea32` — GREEN on both platforms, each green licensed by its own positive control (doyle, 2026-09-12T01:36:11Z).**

Subject: `fold/304-w2-admission` @ `7357ea327c79f5930bdfe5947d427eabd19cf9b8`; windows.rs blob `57e32e52`, linux.rs `88f1489e`, both re-read at each arm rather than carried forward.

| arm | box | command | exit | count |
|---|---|---|---|---|
| Windows (todlando) | HFENDULEAM, fold worktree, own pool | check --workspace --all-targets | 0 | |
| | | clippy --workspace --all-targets -D warnings | 0 | 1m10s |
| | | nextest list -p spt-daemon -E test(bootstrap_firewall) | 0 | population 14 |
| | | nextest run, same filter, unpiped | 0 | 14 run / 14 PASS lines / 0 FAIL |
| | | **control:** `const … u32 = "not a u32"` beside DESIRED_PROGRAM → check -p spt-daemon | 101 | E0308 naming windows.rs:225 |
| | | restore `git checkout --`, blob 57e32e52, porcelain empty → re-green | 0 | 0 err / 0 warn |
| Linux (a) (hertz) | kitsubito, fold sha detached | check --workspace / clippy -D warnings | 0 / 0 | 13 s / 12 s |
| Linux (b) (hertz) | `hertz/304-cells-on-fold2` @ `f94fe044` = 94a04b20 cherry-picked onto 7357ea32 | check / clippy / nextest -E test(/bootstrap_firewall::linux/) | 0 / 0 / 0 | 10 run / 10 PASS / 982 skipped |
| | | **control:** unused binding inside both cfg gates → clippy | 101 | unused variable + dead fn |
| | | restore, linux.rs blob 8264ee9c re-asserted → re-green | 0 | |

Bounds, stated by the arms' own authors and adopted: Linux (a) never compiled FOLD-2's one code edit (windows.rs is cfg-gated) — it licenses only that the Linux product tree still compiles with FOLD-2's prose files present. (b) speaks for the ten Linux admission cells, not the suite. The Windows arm is the ONLY evidence blob 57e32e52 was compiled, and it went red when made to. Two of the fourteen Windows cells assert the constant FOLD-2 documents (`…does_not_satisfy_a_spec_that_wants_no_program_filter`, `…renders_both_halves_and_no_program_filter_by_default`).

Voids and corrections on record, not reds: todlando's first unit arm was killed by him during compilation (exit 137, zero tests executed) after a watchdog printed "stopped" without stopping anything — filed as IR-116, landed on main at `15460d70`. His first "PASS lines 6" was a truncated tail, replaced by the unpiped 14. hertz's control chain reported a wrapper exit 1 that belonged to a trailing `grep -c` printing 0 (IR-107 shape); both clippy runs in it exited 0 and 101 as intended.

Assembly inputs by sha: main `15460d70` + `f94fe044` (carries fold tip 7357ea32 and the cells; bundles verified by sha256 both directions, todlando's independently cut fold bundle byte-identical to mine) + premise `58120cfe`. Dry-run: premise merges clean; f94fe044 conflicts only in docs/INFRA-REGISTER.md, one hunk, both sides appended after the same anchor, main has no IR-105 heading and links to it — keep-both. Assembling now; milestone-batch default holds (no golden until the batch is whole). Decision 2 and Q1 stay with the operator.