What the compiled comparison establishes
All six generated cases pass. The calculation took 57.44 seconds, using six native queries and 162 interval calculations covering 10,368 packet steps. Each calculation retains all 64 packets. Every native position and corresponding first derivative lies inside its computed interval; the maximum outside gap is zero across all 27 parameters and both processing assumptions.
This qualifies the short compiled adapter connection. Its source, inputs, environment, charges and outputs are sealed, and the saved publication passes verification. No aircraft originals or optimizer starts were used. The slower rational-reference controls and known-answer controls independently check higher derivatives; the native reference supplies only first derivatives.
These results do not establish useful full-record uncertainty, all clock and rotation boundaries, native execution error or an aircraft measurement-error law. The long-record study and its width targets remain unchanged.
The nested state reports inherit the reference serializer's dyadic_bits: 96
label. That label does not describe the compiled endpoint encoding. Use the
saved exact endpoints and compiled-kernel metadata: this backend uses outward
binary64 intervals. This metadata issue does not change the comparisons and
must be corrected in a separate successor, preserving this freeze.