# ADDENDUM to PREREG_rushmore_round2 — leg 3a re-run (29 September 2026) Leg 3a FAILED in run 1: worst |n·Kₙ − 1| = 1.11e-05 against 1e-6. That result stands in RUSHMORE2_FISHER_KALMAN_2026-09-29.json. **Diagnosis, a defect of mine.** With the diffuse prior P₀ = 1e12 the first gain is K₁ = 1 − 1e-12. My code updated P ← (1 − K)·P, and 1 − K near 1 keeps only about four significant digits (the spacing of doubles near 1 is 1.1e-16 against a difference of 1e-12). The error of ~1e-4 in P₁ then shows up as ~1e-5 in n·Kₙ. **The leg is unchanged:** same quantity, same criterion (max |n·Kₙ − 1| < 1e-6 over n ≤ 10,000, P₀ = 1e12, r = 1). Only the update changes, to the algebraically identical and numerically stable P ← P·r/(P + r).