Syntonic Nav: A Navigation Filter with No Parameter Table

Pre-registered Simulation against DroneCode Foundation's PX4’s EKF2, with a Fairness Audit of the Rival PX4's estimator, EKF2, sets its gains from 142 configuration parameters (v1.17.0). This note reports a filter of the same kind (error-state Kalman filter over attitude, velocity, position and sensor biases; IMU, magnetometer, barometer, GNSS; in version 2 also optical flow, range finder, two IMUs, two receivers and fault handling) whose every process noise, measurement variance, prior and gate is a statistic of a record of the sensors themselves, read by one declared procedure. No per-vehicle value is set by hand. In steady state, the gain of each scalar channel is the canonical window \(\tau^\star = \sqrt{R/Q}\) of its measured noise–innovation pair (\(\kappa = 1\), the Kalman entry of the Syntony Principle's atlas); what is tested is the Principle's operational claim, measure the pair and do not tune. Both versions were tested in simulation under pre-registration: development seeds, a frozen design with SHA-256 hashes, one run on fresh seeds.- Version 1 held 18 of 18 predictions on 30 seeds. Against the same structure tuned against truth, its tilt and yaw errors were 0.75 and 0.58 of the tuned filter's, with parity on velocity and position, and a drift after a 30 s GNSS outage 1.09 times the tuned filter's, inside its declared band.- Version 2 held 119 of 121 predictions on 20 seeds against PX4's own EKF2, compiled unmodified from v1.17.0 (its 140 EKF unit tests pass) and fed as a Pixhawk 6X as shipped, after an audit that corrected four faults in our feeding of it. The table-free filter diverged in none of seven scenarios (nominal, GNSS glitch, 60 s GNSS outage with optical flow, magnetic disturbance, barometer step, a biased noisy IMU, a dead receiver). On the nominal flight its tilt, yaw, velocity and position errors were 0.46, 0.39, 0.23 and 0.63 of EKF2's, and in every fault its position error during the fault was the lower. Two features of the bench favour it and are stated with the results: the IMU noise is replayed from an X500's raw logs, far above what EKF2's defaults assume, and five of seven sensor models are data-sheet models from which the filter's facts were also read. Nothing here has run on hardware. The next test is the X500 on a three-axis gimbal, flying itself, with one encoder per axis as truth for roll, pitch and yaw, and both this filter and PX4's EKF2 replayed on the same logs. Code and frozen files are withheld until the hardware test and will be deposited as a new version of this record; their SHA-256 hashes, recorded before each confirmation, are listed in the note (Table 5). Authorship and collaboration statement. All prose in this corpus was drafted by large language models under the author's direction and revised in author-directed loops. The author set the questions, chose the hypotheses, designed and ran every experiment, decided every retraction, and designed the corpus as a system (immutable hub-and-spoke record, concept and version identifiers, errata register, hash-frozen firmware, pre-registration gates). Verification is independent of both: the algebraic core is machine-checked in Lean 4/Mathlib, numerical results follow pre-registered protocols, and flight logs are public. Where model-generated reasoning text survives in a published document, it is an artifact of this method and is recorded in the Corpus Status Register. The author alone approves every statement and is accountable for it; no language model is an author. RETURN TO HUB : https://doi.org/10.5281/zenodo.17254395