Shared errors survive the native-model checks

All 24 generated controls pass: the three physical models, both processing cases, and tangent, packet-boundary, broad-calibration and independently generated stationary-motion families. All 162 physical tangent-column checks pass. The largest native center discrepancy is 9.86 nanometres. The study used its 78 charged native evaluations and has finished; its inputs, source and results are sealed.

The calculation now preserves two nonlinear errors common to a recording: initial velocity times the common clock offset, and the initial attitude/lever product. The original 27 physical calibration directions remain. Six additional bounded error coordinates represent those common errors; they are not new physical freedoms or six new choices at each measurement time.

The next check connects all measurement times to one common six-coordinate block, then repeats the complete orientation-cover usefulness comparison. A short native-control pass does not establish useful long-record prediction, whole-program floating-point enclosure, or actual receiver-error coverage. Observed prediction studies remain behind those gates. The 53-stretch model comparison and the protected independent ESA recordings are unchanged.

Sealed results · Frozen scope · Reference-error method · Roadmap.