Common reference-acceleration component controls
Freeze 18 generated identities: three reference-force samples, three nonzero/zero calibration centers and two full physical domains (original yaw range or one original yaw sector). All 27 physical calibration directions and their original limits remain. This component is not a native propagator.
Each identity checks 64 fixed corner/interior parameter points against an independent Rodrigues rotation/engineering-unit force formula. Compare all 27 analytic tangent columns with 54 two-sided point evaluations. Check the three common-error vectors against their declared radii and the exact acceleration polynomial decomposition at four clock center/slope choices, including zero, negative and doubled slopes. Preserve every physical and all 15 auxiliary columns in the expanded 42-column interface. Nonzero rotation, gain, bias, gravity and time centers must use the full formula.
Acceptance: independent acceleration values within 3e-13 m/s², tangent differences within 2e-8 m/s² per normalized parameter, polynomial identity within 3e-12 m, and normalized common vectors within 1+1e-11. These finite numerical checks do not establish a complete floating-point enclosure. Save full vectors, radii, derivatives, bindings and discrepancies; no failed identity may be retried.
Limits: 1,152 probe points / 972 derivative point evaluations / 4,608 polynomial checks, 60 cooperative seconds, one worker with four numerical threads; zero native evaluations, archive reads, observed fits or protected ESA outcomes. Charge identities and reserve operations before execution. No expansions or signals. Freeze sources/environment and deterministic point-generation seeds. Passing permits separately frozen native construction/qualification, not observational use or a longer fitted prediction campaign.