Syntonic Nav: Against Adaptive Filters

A Filter with No Tuning against the Adaptive Filters of Sage-Husa and Mohamed-Schwarz, Pre-registered Simulation Syntonic Nav reads every noise of a navigation filter from records of the sensors and tunes nothing. The classical adaptive Kalman filters make the same promise by another road: they infer their noises in flight from the filter's innovations. This note confronts the two. After a fidelity audit, written and hashed before any duel run, Sage–Husa and Mohamed–Schwarz estimators were embedded in SYN5's own error-state structure, only the noise source changed, in two forms: started from generic defaults (PURE) and started from SYN5's measured facts (INFORMED). One pre-registered confirmation on 20 fresh seeds over 14 scenarios gave SYN5 v4b 42 of 42 criteria against the PURE pair, which diverged in 86 and 107 of 280 runs, and 40 of 42 against the better of all four: both failures were in one scenario, a slowly dragging receiver, against Mohamed–Schwarz started from the facts. Measuring beats inferring from scratch; measuring and then inferring ties with measuring. A follow-up candidate, SYN6, added a raise-only live reading of each receiver's noise. On six held-out scenarios sealed beforehand by a separate agent of the same model it held 33 of 35 criteria, failing where its predecessor would fail too; on the 14 scenarios it held 42 of 42, as did its predecessor and SYN5 v4b, so the earlier failure did not replicate on fresh seeds. The live reading worked as designed and bought almost nothing in position. SYN6 is not adopted; SYN5 v5a remains the flight candidate. The rivals are the founding methods of one family, covariance matching by the innovation; the later variants and the other families of adaptive filtering are not tested here. Simulation only; the code is withheld until the hardware tests are complete. 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. This note is simulation only: no Lean proof and no flight log belong to it; its code and data are committed by SHA-256 and withheld until the hardware tests. RETURN TO HUB : https://doi.org/10.5281/zenodo.17254395