nx_rankmap.nx

nx_rankmap.nx -- ONE canonical ranking vocabulary, anchored to the INDUSTRY standard. OPERATOR 2026-07-31: "why do we have s class and exceed i dont understand the rankings ... we need the ranks researched as of july 2026 and matched to the industry as terminology and all our terminology interoperable so our capabilities and benchmarks and oracles have clarity on where they rank and cant be gamed". THREE DEFECTS THIS NAMES AND FIXES: (1) TWO LIVE LADDERS, same 0-5 shape, different rungs: ecomat : ABSENT / TOY / FUNCTIONAL / PRODUCTION / S-CLASS / EXCEED capgraph : ABSENT / TOY / WORKS / GATED / ADOPTED / SOTA Two answers to "what level is this" is the same two-sources-one-name defect as two roots, two trees, two config files and two clocks. (2) MIXED AXES. ecomat 0-3 are CAPABILITY properties (what the thing does). 4-5 are COMPARATIVE CLAIMS (triangulated parity / win vs a NAMED best-in-class). One scale carrying two different questions is why the top rungs read as incomprehensible next to PRODUCTION. (3) GAMEABLE -- now closed at the source. permil_to_level used to award 4/5 from a COVERAGE PERCENTAGE (p>=700 -> 4, p>=900 -> 5), so a high-coverage census minted "parity with gcc/llvm" with NO COMPARISON EVER RUN. Capped at PRODUCTION 2026-07-31. THE INDUSTRY ANCHOR (researched July 2026): TRL 1-9. NASA-origin, adopted for DoD procurement, ISO-canonized 2013, with MRL/SRL siblings. TRL measures READINESS ONLY -- it deliberately says NOTHING about beating a competitor. That separation is exactly what we were missing. ★★★★★THE CLARIFYING RESULT: S-CLASS and EXCEED MAP TO THE SAME TRL (9). They differ ONLY on the comparative axis. Two rungs that share a readiness level and differ only by comparison were never one scale -- which is the structural proof that the axes must be split, not merely renamed. license_tier: ORIGINAL No hw writes (Rule 26).