What the original-observation probes establish
Independent exact arithmetic confirms 6/12 candidate calibrations fit the original five perturbed training observations within ±1 metre. All twelve points, scales, forward outputs and charges are checked; failed training matches remain excluded from the spread. No trajectories are rerun.
| Processing | Accepted calibrations | Sampled N spread | Sampled E spread | Sampled height spread |
|---|---|---|---|---|
| Zero removal | 4/6 | 13.328239 m | 5.365217 m | 0.641800 m |
| Profiled removal | 2/6 | 5.898277 m | 6.049479 m | 0.127705 m |
These sampled spreads do not exceed the original 20/20/50-metre targets. This test therefore does not establish that the targets are impossible for the original observations. It also does not prove them achievable: unexamined calibrations can produce a wider range. A useful outer enclosure remains necessary.
The earlier zero-reference fixture admitted 21.75 m north separation. Its observations and expansion were different. Neither its larger spread nor the smaller spreads here can be transferred between the problems.
The audit uses the actual stored observations, including their perturbations, rather than silently substituting the known truth or zero expansion. It independently recomputes nonlinear feasibility, verifies the LP primal directions and checks every fixed candidate construction. It does not assert exact optimization, native execution-error coverage, an aircraft error law or statistical significance.
The original long-record study continues unchanged. Its earlier wide-range failures remain failures; these sampled points do not replace its required whole-domain prediction bounds. The 53-stretch findings are unchanged.