What the timing search established
The matched search completed all 38 starts and eight exact reconstruction checks in 6.59 hours. 37 individual fits passed; both fresh controls passed. Only two of six selected independent-repeat comparisons passed. The four-minute IMU21 horizontal-only sensitivity result therefore remains unresolved.
Each model received the same three timing starts and two independent parameter seed families. The lowest returned cost in each family determined its result; no inconvenient attempt was discarded. This is a finite search, not a guarantee that the global optimum was found.
| Comparison | What happened |
|---|---|
| Specified-field globe, both processing cases | Both independent searches found the same lower-cost solution and passed the cost and prediction checks. |
| Zero-field globe, both processing cases | Different local solutions remained. Selected costs differed by about 0.0145%; north/east prediction RMS differences were about 3.0/2.0 mm, above the 1 mm limit. |
| Flat, no upstream removal | Costs agreed within the limit, but north prediction RMS differed by 1.32 mm, above the 1 mm limit. |
| Flat, profiled removal | Selected costs and predictions disagreed: north/east RMS differences were about 38/62 mm. |
The selected specified-field globe costs are about 0.0807, zero-field globe about 5.56, and flat about 1,774–3,646. That ranking persists across the searched seed families. The numerical gate still withholds a complete matched interpretation because the other candidates did not reproduce their own best predictions. An unfinished or nonrepeatable fit does not provide evidence for flat geometry.
Why height remains important
Even the repeatable specified-field solutions have about 646 m of unscored height error. Other selected target solutions have roughly 20–25 km. These are reconstructed errors against receiver altitude, not actual aircraft motion. Removing height from the objective permits trajectories that are poor full-flight reconstructions. The earlier complete IMU8 sensitivity comparison also has large unscored height errors.
The check shows how the horizontal score behaves when height is not scored. It does not remove height from initialization or propagation, isolate gyroscopes from accelerometers, or establish a physically adequate flight reconstruction. The original full-position 53 globe / zero flat / zero unresolved findings remain separate and unchanged.
Next measurement check
Further widening this timing search is deferred. The next step defines what replaces a removed IMU channel and checks that replacement on known-answer examples before any new recorded-data fit. Gyroscope-free attitude cannot be silently supplied by the fused navigation solution; acceleration-free tilt cannot still be initialized from the omitted accelerometer observations.
All attempts and source metadata · Measurement-contribution plan.