todlando -- FOLD-1 EXECUTION, END 2026-09-11T12:26Z. Open 12:22Z, ~4 min against a 600 s cap. FIVE arms, all ran, no holes. SUBJECT 61d3459fd6db5be0e66951b9f6d2a05822865518 throughout; every number carries that sha.

ARMS.
  F1 u1-R  RED AT THE PREDICTED LINE AND VALUES. exit 100, "1 test run: 0 passed, 1 failed, 991 skipped".
     windows.rs:482 -- assertion `left == right` failed: the spec carries the BOUND port
       left: 5470    right: 56025
     desired_spec ignoring its bound_port argument is caught at the equality. The cell is not a tautology.
  F2 u2(ii)-R  RED, and at the assertion I NAMED IN ADVANCE. exit 100, "1 test run: 0 passed, 1 failed, 991 skipped".
     windows.rs:506 -- "the working default carries no program filter"
     That is `assert!(want.program.is_none(), ...)`, the cell's FIRST assertion, exactly as filed.
  F3 hertz's SIX  6 tests run, 6 passed, 986 skipped. Zero-match gate: 6 == 6, SATISFIED. His cells are unchanged by this commit and I touched none of them.
  F4 my FOUR  4 tests run, 4 passed, 988 skipped. Gate: 4 == 4, SATISFIED.
  F5 unfiltered  992 tests run, 992 passed, 0 failed, exit 0.
     `cargo nextest list -p spt-daemon --lib` at the SAME sha: 992. RUN == LIST, both 992, both at 61d3459f. No number from another tree entered this prediction (the W-1 lane's 981 has no standing here and I did not cite it).
  F6 pool  no SPT_POOL_FOREIGN, no lane-identity refusal. Pool 5.54 GiB.

A LIMIT IN MY OWN ARM DESIGN, NAMED BECAUSE IT IS EXACTLY THE CLASS THIS FOLD EXISTS TO KILL:
  F2 proves the POLICY CONSTANT is load-bearing. It does NOT prove the PREDICATE's program comparison is falsifiable. The mutation stopped at the cell's first assertion, so `spec_satisfied_by(&with, &want)` was never exercised under it. The arm that would test that is a mutation INSIDE the predicate -- force `program_ok = true` -- and I did not run one. So: constant wired, comparison asserted only by the green path.
  Same shape holds for the other three cells: each is green, and only F1's subject has been shown red-capable. I am not claiming the seam is fully falsifiable; I am claiming two of its arms are, by measurement, and naming the rest as unproven. If you want the predicate arms, they are three more mutations in one window and I will file predictions for each.

THE LEAKY SET CHANGED, AND IT REFUTES A CHARACTERISATION -- INCLUDING ONE I REPEATED.
  FOUR leaky here, not five: brainproc x3 (clear_before_spawn, ready_but_old_gen, stale_generation_minus_one) plus broker::tests::windows_session_is_zombie.
  `brainproc::tests::trial_kills_alive_never_ready_candidate_before_rollback` DID NOT LEAK. I checked it is not an absence artifact before saying so: the cell exists in this tree, and it RAN -- "PASS [0.559s] (116/992)" -- it simply did not leak.
  So hertz's "the brainproc FOUR leak in every run -- cell-shaped" now reads 3-of-4 cell-shaped and one run-shaped across four runs on three trees. I carried that characterisation into my own W-1 END without re-deriving it, which is the inherited-claim habit again in its mildest form. IR-108 should record the count as measured-per-run rather than as a set.

REVERTS, both proven by oid AND byte shape, on a CRLF tree:
  pre-arm: 26514 bytes, 567 CRLF, 0 bare LF, work == head == f634ff2065de6f704b07f94f33fe642c50a23d63.
  after F1 revert: identical on all five. After F2 revert: identical. Final: identical, tree CLEAN.
  Anchors were rebuilt from the file in hand and translated to CRLF before counting; both counted exactly 1 at preflight, so neither arm mutated blind.

TRACEABILITY: checker 0.4.1, tree pins WANT=0.4.1 -- instrument MATCHES, first time this milestone, so no IR-104 hole to declare on this lane.
  REQ-WEB-LAN-BOOTSTRAP-FIREWALL: required [doc, impl, unit], reads +doc +impl +unit -int. The folded int clause is present and OFF.
  check exits 1 with NINE [must] misplaced_tag, all code-stage tags in docs/ prose, zero naming bootstrap_firewall. YOUR NOTE TAKEN: they are fixed on main after this base, so asm reads zero -- that resolves them as a base artifact rather than a debt this lane carries, and I will not re-report them as findings at assembly.

BOX IS FREE FROM ME: post-census positive control TRUE, ZERO survivors across builders and the lane's own test binary. Free 142.4 -> 130.4 GiB (the fold pool is a fresh 5.54 GiB tree; the rest is the two cold builds this lane paid for).

NEXT, on your word already given: the three Linux seam riders (backend_from, parse_ufw_rules, ufw_preflight taking &[UfwRule]) as ONE commit on top of 61d3459f with its own compile-gate, no execution here because they are cfg-invisible on this box, no Linux cell written by me. Starting them now.