todlando -- PREDICATE ARMS, START 2026-09-11T12:28Z, on doyle's word. SUBJECT 61d3459f, lane fold/304-w2-admission, tree CLEAN, pool todlando-304-fold. Box free (hertz on kitsubito). Cap 600 s. First act of the window.

BOTH CELLS DDoyle ASKED ABOUT EXIST -- no finding, but they live in ONE cell, which matters for reading the reds: an_unrestricted_remote_does_not_satisfy_a_narrowed_spec carries BOTH the wrong-remote assertion (:538) and the wrong-profile assertion (:553).

THREE MUTATIONS, ALL INSIDE spec_satisfied_by, the constants untouched:
  P1 :214  `if !program_ok {` -> `if false {`
     PREDICT RED in a_program_bearing_rule_does_not_satisfy_a_spec_that_wants_no_program_filter at :513
     (`!spec_satisfied_by(&with, &want)`), NOT at :506 -- the constant stays false so the first
     assertion still passes. This is the arm my FOLD-1 END said was missing.
     Second-order, stated so it is not read as a surprise: :525 would also fail under this mutation,
     but :513 fires first and is what the runner will name.
  P2 :217  `if observed.profile != want.profile {` -> `if false {`
     PREDICT RED at :553, the LAST assertion of the remote cell (`wrong_profile`). The three earlier
     assertions in that cell are remote-shaped and survive a profile mutation, so a red anywhere
     ABOVE :553 would mean the profile check is load-bearing somewhere I have not accounted for.
  P3 :224  `seen == wanted` -> `true`
     PREDICT RED at :538, the FIRST assertion of that same cell (`unrestricted` becomes satisfied),
     so P3 and P2 redden the SAME cell at DIFFERENT lines. If both reds land on the same line my
     mutations are not independent and the arms prove less than they appear to.

Each mutation reverted immediately after its own arm; revert proven by oid AND byte shape (CRLF tree, 567 CRLF / 0 bare LF, blob f634ff20). Anchors rebuilt from the file in hand and counted at preflight; any count != 1 aborts that arm rather than mutating blind.

ZERO-MATCH GATE on every arm: exactly 1 test run, or it is a NON-MEASUREMENT.

Then, same window or the next on your word: the three Linux riders (backend_from, parse_ufw_rules, ufw_preflight over &[UfwRule]) as ONE commit with its own compile-gate, no execution here, no Linux cell written by me.