Methods × implementations — the complete grid

One row per method × problem size × thread count; one column per implementation (C++ build tiers, every language binding, and the canonical external references). Each cell pairs a value with a parity verdict — reference gate for the C++ core & external libraries, binding gate for the language bindings. Switch to divergence to read ΔRMSE (numeric methods) and Jaccard set-overlap (selectors). NR (not run) and n/a (not available) stay distinct and are never coerced into a number. Prefer a guided view? Open the card explorer.