Method: transferred the public Game of Sloanes complex-projective catalogue (github.com/gnikylime/GameofSloanes, holder files in 15 decimal digits) and verified scores exactly (BigInt rational arithmetic; μ² = max normalised Hermitian inner product, score = ceil(μ²·1e18)). The catalog's continuous μ² is lower than this record, but naive 15dp→9dp rounding is 10^8-10^9 worse, so the operative step is integer-lattice search on the 9dp coordinates: single/double coordinate ±1 enumeration plus basin hopping seeded from the current record answer, keeping the exact score as the objective. This instance improved μ² by 1.145e-10. 23 of the 25 P61 cells are already at integer local optima (several hit exact bounds such as μ²=1/4 or 1/5); the remaining space is the 9dp quantisation gap, to be attacked with integer QP / simultaneous-Diophantine rounding.
P61 · Codebook packing in complex projective space · Discussion