Duration growth before longer prediction comparisons
The complete shared-reference comparison clears its short improvement target, but still permits roughly 31 m at 1.8 seconds. Screen the remaining nonlinear allowance before launching a more expensive complete prediction comparison. This is a generated propagation diagnostic, not another aircraft model fit.
Freeze all three candidates, both processing cases and all eight original yaw sectors: 48 identities, each with 400/800/1,600-packet prefixes and query times 1.8/3.6/7.2 seconds. Extend the existing deterministic input formulas; verify their first 400 packets, configuration and truth against the completed short study. Do not replace the original 27 bounds with a smaller fitted box. Use the same shared-reference construction and retain all six common errors.
For each prefix, save a native center and all 27 tangent columns, one global bound, and three independently evaluated parameter probes: both domain corners and the fixed interior probe used by native qualification. Require native/bound centers within 1 micrometre, actual common errors within their normalized bounds (1e-12 comparison tolerance), and factored probe errors within the unshared allowance plus 10 micrometres. Apply identical checks to every model/processing case; preserve failures and unsupported regions.
Report the unshared allowance's north/east/height contribution, its position and lever parts, common-error radii and growth with duration. These are conservative calculation terms, not measured instrument noise or minimum attainable physical uncertainty. This diagnostic does not calculate training-feasible regions or qualify full prediction usefulness. No criterion is tuned to a future receiver value; no original files, observed fits or protected ESA outcomes are read.
Limits: 576 native calls / 537,600 native packet visits, 144 bound calls / 134,400 reserved bound steps, 900 cooperative seconds, one worker with four numerical threads. Reserve before every operation; no retries, allowance expansion or process signals. Freeze source, inputs and environment; verify the passing short comparison before launch. A correct but rapidly growing result calls for a mathematical refinement, not a larger aircraft fitting campaign. Whole-program roundoff and observational coverage remain unqualified.