nx_frame_score.nx

nx_frame_score.nx -- score ANY captured frame with the hardened nx_game_critic, ONE ruler ONE scale. WHY (seq890/seq1004): the head-to-head vsbench contract says a Tier-3 EXT cell must be measured off a captured artifact by OUR instrument -- but the organ that minted the first EXT cell existed in NO source tree and was never registered, so the evidence path was not reproducible by any agent. This is that organ rebuilt from the first byte as a REGISTERED MCP tool: OUR render and a REFERENCE render are graded by the SAME deployed binary on the SAME scale, on the NAS, with the artifact on disk. INPUT FORMAT NXFH1 (text-hex framebuffer -- deliberately TEXT so it travels the fs-write lane safely; raw binary through JSON-string tooling is the proven corruption path): line 1: "NXFH1 <w> <h>\n" then w*h pixels as 6 lowercase hex chars each (rrggbb), any whitespace between pixels ignored. Encoder: scratchpad/png2hex.js (same code path for BOTH sides). USAGE: nx_frame_score <path.nxfh> | nx_frame_score selftest Composes nx_game_critic (gc_score/gc_quality/gc_verdict) -- nothing re-implemented. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0