Eight-minute inputs with explicit clock uncertainty
The earlier representative prediction plan covers 438 seconds. Four-minute inputs cannot qualify that complete history. This separate software preparation extends the existing analytic generated-input recipe to 96,000 packets at 200 Hz, about eight minutes. It preserves the original 48,000-packet prefix bit for bit, including signed values, periods and all physical configuration. No aircraft measurements are extended, interpolated or changed.
Declare a new clock-offset limit of ±1 second. All other physical calibration limits, initial conditions and processing assumptions stay as in the original broad cases. This clock range is a qualification assumption, not a measured instrument specification. Preserve the old fixed-clock studies unchanged.
Component and API checks
First check the input recipe, prefix, duration, retained calibration columns, clock declaration, provenance and invalid-input rejection with cheap deterministic tests. Use a supplied parent artifact hash; never label new synthetic suffixes as measurements from the parent file.
Then compare the old and current native argument interfaces for all three candidates and both processing assumptions. Query 438 seconds at a normalized clock value of −1, and 478 seconds at a normalized clock value of +1. The second point also changes the other calibration coordinates by the same fixed small pattern used in the earlier compatibility study. Both queries retain full ±1-second clock support within the new input. Compare positions, all 27 Jacobian columns and warning bits exactly; require a nonzero clock column.
Both callers use the same compiled kernel. Agreement tests the interface and duration/clock handling, not independent physics or a nonlinear uncertainty bound. Generated positions from this kernel must not become independent physical truth.
Finite allowance
Six identities, four native calls each, 24 calls / 2,304,000 maximum packet visits, zero interval-bound calls, optimizer starts or aircraft reads. One worker and one numerical thread. Estimated two to four minutes, with a fixed ten-minute cooperative ceiling and no retries or signals. Freeze the generated suffix, parent hashes, complete sources and numerical environment before launch. Failed cases remain failed.
Success makes complete-horizon software inputs available. Useful bounds derived from their own training observations, an adequate shared observation-error law and supported numerical treatment still precede new aircraft prediction fits.