Duration growth is much smaller with the common acceleration preserved
All 48 identities pass at 1.8, 3.6 and 7.2 seconds: every physical model, processing case and original orientation sector. The generated inputs are byte-identical to the previous duration screen. Native centers, all physical directions and known parameter probes retain the original comparison limits.
The largest unshared width contributions are:
| Duration | North | East | Height | Previous construction, largest axis |
|---|---|---|---|---|
| 1.8 seconds | 0.105 m | 0.131 m | 0.071 m | 15.42 m |
| 3.6 seconds | 0.355 m | 0.411 m | 0.231 m | 45.26 m |
| 7.2 seconds | 1.031 m | 1.164 m | 0.815 m | 150.23 m |
Preserving the common acceleration removes most of the earlier overestimate. These numbers are contributions to the mathematical allowance, not complete training-constrained prediction widths, measured aircraft errors, sensor drift or calibrated intervals. They cannot be compared directly with the 1.8-second complete union widths as though both tables measured the same quantity.
The study completed in 1,918 seconds, with 576 charged native calls and 144 bounds. Source/input/environment hashes and retained outputs are verified. No archive originals, observed model fits or protected ESA outcomes were read. The existing 53-stretch empirical result is unchanged.
The next scientific gate is a longer complete prediction comparison, including the joint training constraints and every original orientation sector. The separate execution qualification prepares four workers with one thread each for this more expensive work. One-thread native compatibility must pass before using that new allocation. Recording-scale usefulness, numerical treatment and actual receiver-error adequacy remain open.
Complete results · Frozen scope · Earlier duration screen · Complete short comparison · Roadmap.