Geographic propagation and receiver-position bounds
All 72 generated controls passed. This closes the single-packet geographic predictor/corrector and antenna-position calculation around the unchanged native integration code. It adds no aircraft-data fits or new scientific detections.
The calculation preserves all 27 calibration, initialization, lever-arm and timing directions. Cases cover both hemispheres, all three physical candidates, narrow and broad calibration regions, and processing fractions centered at either physical boundary. Timing is fixed at a packet endpoint or varies strictly inside a packet. Continuous longitude is preserved.
There are 4,608 declared probes. Values are checked against the original C kernel; all parameter derivatives are checked against the original C++ tangent and independent complex-step calculations. The largest derivative discrepancy is below 2.85e-14 metres per normalized parameter. The analytic remainder bounds cover their declared regions; sampled discrepancies are checks, not the source of those bounds. Comparisons below the stated arithmetic tolerance are recorded separately, rather than presented as precision measurements.
The study used 4,680 C calls, 72 C++ calls and 6,624 independent references in 13.94 seconds. All 246 publication artifacts pass their SHA-256 checks. The frozen plan and result tables preserve the exact scope.
What remains
A short-step bound need not remain useful after thousands of repeated steps. The next component must propagate complete histories and handle receiver-time uncertainty crossing packet boundaries. Whole-record arithmetic and shared receiver-error assessment also remain open. These checks precede a new withheld prediction study; earlier aircraft findings and failed studies are unchanged.