crossbias --selftest: OK - the closed-form chord bias matches the numeric interpolation in sign, magnitude and its vanishing at both interval ends; the seven crpin4m crossovers this lane quotes (396.4/415.3/398.0 e8, 396.4/412.5/401.2/415.7 p4) reproduce and the e8 fine ladder stays a BOUND in wall; crband m=384 is still the 30%-floor outlier the cband refusal rests on; the non-monotone guard rejects a pair on cpp4f; and quad_root finds an exact root and refuses fewer than three points, and the pure-noise second-difference sign-disagreement rate is still 73.2%; and arm 5 still prints its published table (3.9/4.2/5.3/0.8 rows per one per cent, 3.4/3.6/1.3/0.2 rows of rung noise) through the ONE shared w3winred.rung_noise