# ADDENDUM 2 to PREREG_rushmore_round2 — leg 10b, second re-run (29 September 2026) Run 2 (with addendum 1's root-finding) FAILED identically: worst 5.10e-05 rad, and at |R| = 5.2164. **So addendum 1's diagnosis was wrong.** The slope near |R| → 1 was not the cause. Run 2 stands in RUSHMORE2_CROWDS_run2_2026-09-29.json. **The real cause, found by replaying the failing crowd.** The liar's aim (0.2009) is inside the reachable cone [−0.0303, 0.3555], so the true best distance is 0. The optimal report is β = 3.141357, just below +π, where my grid of [−π, π] wraps. The grid's minimum was at index 0 (β = −π). My search window and my Brent bracket both looked only at indices 0–2 and never crossed the seam to +π. A defect of mine, in my numerics. **The leg is unchanged:** same crowds, same formula, same criterion 1e-6 rad. The measurement now treats the report angle as circular. The grid is [−π, π) without a duplicate end. Every sign change of ψ(β) − aim around the whole circle, the seam included, is root-found with brentq. Where there is none, bounded Brent runs on a bracket centred on the grid minimum, taken circularly.