Smaller calibration rays in the complete nonlinear model

The earlier twelve linear-extreme proposals all failed nonlinear training. This study checks whether smaller versions can fit training positions yet separate later predictions. It uses the same two generated specified-field histories and both processing assumptions, not aircraft or protected ESA data.

Keep all twelve saved proposal directions, all 27 parameters, the original physical limits and ±1-second clock support. For each direction use only the 292 saved training epochs, its zero-reference Jacobian and its earlier nonlinear endpoint. Write L for the linear training change and Q for the remaining endpoint change. Choose a dyadic scale by 48 fixed bisections, requiring max|L|*a + max|Q|*a² <= 0.75 m. This quadratic expression is a proposal heuristic, not an enclosure of the nonlinear trajectory.

Save each represented scaled point before one new full nonlinear query at all 293 original times, including the future endpoint at 438 seconds. No new outcome may change that point or cause a retry. The actual training-feasibility criterion remains ±1 metre on every coordinate at all 292 earlier epochs. Neither the future outcome nor its ranking determines feasibility.

Reserve one numerical thread and two for the running long-record study, with a ten-minute cooperative ceiling. Freeze source, generated-input references, environment, compiler and counts before launch. No optimizer starts, original aircraft reads, process signals, retries or allowance expansion.

Report all outcomes. Among numerically feasible points and the zero reference, report the spread of future predictions under each processing assumption, beside the unchanged 20/20/50-metre width targets. This is a numerical ambiguity diagnostic, not a complete uncertainty enclosure or a proof of native floating-point execution error. No feasible sampled separation would prove unique prediction either. The original long study has perturbed training data; this separate zero-reference fixture must not be described as its result.