START — #304 premise-line follow-up lane, window 1.

GRANTED BY: doyle, msg 7XIPZX6K, 2026-09-11T12:06Z.
Verbatim grant: "EXECUTION WORD for test/304-premise-lines: todlando's W-1 ENDed
12:06Z with a post-census whose positive control read TRUE and zero survivors;
the box is yours from your compile-gate exit. Arms: base green x2, tip green x2,
then HERTZ_RIG_TMP_IN_REPO=1 red x2 AT THE PREMISE LINE. Cap per your START.
Post-census after."

Compile-gate exit that opens the window: cargo check -p spt-daemon --lib --tests
at 58120cfe, "Finished dev profile in 49.45s", spt-daemon checked. The gate was
proven to SEE the new lines by a deliberate red (both premise asserts renamed to
a non-existent fn; E0425 at registryhost.rs:1376 and projwriter.rs:806, "could
not compile spt-daemon (lib test)"), then reverted; 0 NEGATIVE_CONTROL survivors.

SUBJECT
  worktree C:\Users\decid\Documents\projects\spt-core\.worktrees\hertz-premise
  branch   test/304-premise-lines
  base     4b54a95d  (origin/main, fetched and confirmed this session)
  tip      58120cfe  test(daemon): assert the outside-any-repo premise ...
  pool     <worktree>/target, claimed label hertz-premise, base 4b54a95d22d9
  rig      .spt/preserved/hertz-premise-58120cfe (forked from the W1 rig;
           POOL and TREE are env-driven and REFUSE when unset)

CAP
  6 arms, one cell each, 900s deadline per arm.
  tip-registryhost, tip-projwriter          expect GREEN
  ctl-registryhost, ctl-projwriter          expect RED at the PREMISE line
  base-registryhost, base-projwriter        expect GREEN
  No other producer. No golden. No execution beyond these six.

PREDICTIONS, FILED BEFORE THE RUN
  1. Both cells GREEN at base 4b54a95d and GREEN at tip 58120cfe. The premise
     line is an assertion that must HOLD on a correct rig, so it changes no
     verdict on a correct box.
  2. Under HERTZ_RIG_TMP_IN_REPO=1 both cells RED, and RED AT THE PREMISE LINE:
     registryhost at the new assert, NOT at the old vec equality; projwriter at
     the new assert, NOT at the old "proj-a" display assert.
  3. doyle's caution, taken as a live outcome and not a footnote: the env var is
     read by the RIG (lane_run.priv_for, PARENT env), not by the cells. A RED at
     an OLD line under the control means the rig moved TMP but my assertion sits
     after the derivation it guards — the premise line is then MISPLACED and I
     move it rather than argue the result. I name the line either way.
  4. Any arm whose nextest summary does not read exactly 1 test run is a
     NON-MEASUREMENT, not a pass: a zero-match filter exits 0.
  5. Any control arm resolving the SAME tmp base as its paired tip arm is a
     NON-MEASUREMENT: the switch did not reach priv_for (window 6's defect).