The credit descent as a rail — the pass design/plans/tng-intro.md §2 says
jags, and the jag is what a straight line refuses.
Fitted on the screen residual over its own measured beats, with f2050 and f2065 left out of the fit. Those two are where the refit splices the box channel to the Bussard-cap channel and the authored x steps 0.126 of the frame between them — 0.368 to 0.483 and back to 0.357 over twenty-five frames. No straight line contains that step, which is the finding: the heading swung 82.0° off the nose at f2068 because of it, and that is why the old property test's descent window stopped at f2036.
The unfitted check is the entry. A straight pass appears at the projection of its own reversed direction, and this one's is (0.454, −0.000) against a measured f1775 mark of (0.462, 0.005) — eight thousandths of the frame, from a number nothing fitted to it. Projected back across the rest of the pass the rail holds the measured track to 0.05 of the frame or better everywhere except f1920–1960, where it runs 0.09 wide; the reference's own descent fits a line at 5.0%, so that is the material and not the model.
SHIP_RETURN keeps every one of those beats. They are the specification the
rail is checked against, not the path it flies —
design/plans/tng-intro.md §3.3 — and they still drive the skim, which
begins where this leg ends.