Local Level Lab

Airborne measurements. Open methods.

Packet-boundary check stopped the first run

All six starts ended after their first native query, before any source bounds or predictions were calculated. The run charged six queries in 19 seconds. It says nothing about whether the intended prediction ranges are useful.

The queries themselves succeeded. The wrapper incorrectly treated a nonzero packet-boundary counter as a motion warning. In the native kernel, that counter records a query at the start or end of an IMU integration packet. It is separate from the failure status. An independent timestamp calculation reproduces all six saved counters; a native-kernel regression checks boundary and interior queries separately.

The original failed results and frozen acceptance rule remain unchanged. The separate corrected study checks the counter against timestamps, retaining the same input records, calibration bounds, query times, width targets and finite per-case budgets. This correction does not establish smooth derivatives at packet boundaries or full numerical coverage; those remain explicit qualification requirements.

Preserved results · Roadmap