Timing derivative and path results
Completed 29 fixed evaluations in 5.28 minutes, with zero optimizer starts. All primary and repeat centers reproduced exactly. No fit, repeat or scientific labels changed.
Independent timing derivatives
The analytic timing column is compared with symmetric one-microsecond residual differences. Errors below are relative vector-norm discrepancies, not statistical uncertainty. Distance to the nearest nominal integration boundary is reported separately.
| Processing / endpoint | Relative derivative error | Nearest clock boundary (µs) | Perturbation crosses boundary? |
|---|---|---|---|
| unsubtracted / primary | 8.30245e-05 | 709.089 | no |
| unsubtracted / repeat | 7.76991e-05 | 1750.231 | no |
| profiled_removal / primary | 8.7083e-05 | 709.088 | no |
| profiled_removal / repeat | 7.86739e-05 | 1750.231 | no |
| IMU8 control / primary | 0.000149568 | 1368.381 | no |
Full-vector path
Every nuisance coordinate moves together along the straight line between saved endpoints. These are fixed-point costs, not reoptimized timing-profile costs. The timing column gives its parameter value relative to the recorded receiver times.
| Fraction from primary to repeat | Timing (ms) | Unsubtracted cost | Profiled-removal cost |
|---|---|---|---|
| 0.000 | -84.227688 | 5.564176204 | 5.564176206 |
| 0.125 | -86.597545 | 5.564913593 | 5.564913591 |
| 0.250 | -88.967402 | 5.566181738 | 5.566181738 |
| 0.375 | -91.337260 | 5.567721408 | 5.567721407 |
| 0.500 | -93.707117 | 5.569500712 | 5.569500713 |
| 0.625 | -96.076974 | 5.571065595 | 5.571065594 |
| 0.750 | -98.446831 | 5.572399014 | 5.572399015 |
| 0.875 | -100.816688 | 5.573230579 | 5.573230577 |
| 1.000 | -103.186546 | 5.573073571 | 5.573073571 |
Interpretation
The five derivative comparisons differ by 0.0078–0.0150% in vector norm. No perturbation crosses a nominal integration boundary. This finds no gross timing-derivative fault. It does not certify the accuracy of a nearly cancelled objective gradient or prove either endpoint is globally optimal.
Both processing cases give almost identical sampled costs. The path rises from the primary toward the repeat, reaches a sampled high point just before the repeat, then falls slightly. This is consistent with distinct locally stationary solutions connected through coupled timing/calibration directions. The path was not reoptimized at intermediate times, so it is not a minimized timing profile.
No failed repeat passes from these points. The horizontal-only fits retain their large unscored height errors; the original full-position 53 globe / zero flat / zero unresolved results are unchanged. The next task is known-answer qualification of a deterministic multi-start method before any new observed fits.
Recovery qualification plan · All saved point metrics · Frozen scope · Preceding diagnosis.