Measure transport improvements against the actual saved bounds
The coordinate-transport candidate has passed all 24 native controls. Before adding the new body-source interval or running a complete prediction campaign, measure its effect on the six existing sector-4, 7.2-second generated histories. The saved combined-width target still fails in both east and height for the specified-field groups; a vertical-only component improvement is insufficient.
Reconstruct each original typed training-conditioning certificate, checking its six exact supports without a new LP. Its binding must match the conditioning in the saved source bound. Use that bound's original physical domain, source, reference point, time and all 51 physical/shared columns. Evaluate the separately qualified cached transport candidate once per identity. Preserve exact reference centers, coefficient matrices, common-error bindings and conditioning identity.
Report old, new and per-axis minimum unshared error allowances in metres. The minimum is valid only after the reference maps match exactly. Keep all unimproved axes visible. This is an actual-baseline preflight, not a complete sector union, calibrated coverage assessment or scientific decision. It does not itself qualify a position-range target.
Finite allowance: six identities, six setups, 36 exact support verifications, six propagation bounds and 19,200 bound packet-steps (two native passes per 1,600-packet history). No native trajectory evaluations, LP proposals, optimizer starts, original-aircraft reads or protected ESA data. Use one worker with four numerical threads; each bound retains the original prediction study's 120-second cooperative limit, with an overall ceiling of 900 seconds. No retries or budget extensions. Freeze sources, inputs and environment before launch.
Use the results to choose the next method-development step. Retain all earlier failures, and do not start another complete prediction comparison merely because the local source interval is narrower.