nx_forkgrade_gate.nx

nx_forkgrade_gate.nx -- CE8 (codeeffectiveness fg_trend): does the fork census leave a TRAJECTORY, and only where it should? END-TO-END: forks the promoted nx_forkgrade (or the <subject_elf> given as argv[1], absolutised against the CWD) over a THREE-FILE fixture pair it plants under /tmp -- identical, A-superset, bidirectional -- so the census has known answers, then asserts the ONE new capability: every successful census appends one row to a spine, announced (spine= fd= wrote= of=), and two censuses in time leave two rows. The spine path is DERIVED from dirA, so the fixture writes a SIDECAR beside its own root and the PRODUCTION spine (knowledge/status/forkgrade.spine) is asserted UNCHANGED by size: a trial run can never forge the production trajectory. Every expected count below was decided from the fixture before the run. RED against the pre-fg_trend binary by construction (no spine line, no sidecar): the RED-before-GREEN this gate owes. The subject forks nx_srcdiff.elf by BARE NAME from its CWD, so the run is started in the directory where that ruler resolves (CWD or its parent) -- derived from data, never a pinned absolute path; absent ruler = SKIP, never RED. exit: 0 GREEN . 1 RED . 3 SKIP license_tier: ORIGINAL No hw writes.