Returns base offset of node n's nc-wide row in a flat [N ? nc] array.

v1.8.2 A2 encoding (POST_OPTIMIZATION_REVIEW Sec 4.3): the showdown kernels now read a 1-byte `category` per cell (pre-thresholded at precompute time)

v1.8.2 A2 encoding: takes a pre-thresholded category byte per cell rather than re-bucketing the continuous ev. See cpu_simd.h header + cpu_kernels_avx

v1.8.2 Phase 2 (kernel-level batching). Processes M terminals in a single kernel call, all sharing the same matchup table. Key restructuring: loop ne

v1.8.2 Phase 2 (kernel-level batching): scalar reference. Numerically equivalent to calling showdown_oop_full(...) for each terminal in turn — the AVX

Solve endpoint — accepts JSON request, returns solver response. Context protection: never returns full 1326-combo matrix, only aggregated data.

Returns true if compute_combo_evs_gpu() / compute_best_response_gpu() produce meaningful results. Solver checks this before dispatching.

POSTFLOP_STYLE convergence parity. Unlike STANDARD, the postflop schedule converges to sharp, less-mixed equilibria and is float-sensitive in the earl

v1.8.2 Phase 2 parity: showdown_oop_full_batch must produce numerically equivalent output to running showdown_oop_full once per terminal. Tests both s