nx_mfrow.nx

nx_mfrow.nx -- UPDATE ONE ROW OF A TREECANON MANIFEST, AND NOTHING ELSE. WHY THIS EXISTS, measured 2026-08-16. The treecanon family has a generator (nx_treediff, nx_treehash), two comparers (nx_treediverge, nx_hashdiverge) and a gate that refuses (nx_treecanon_gate). It has NO surgical write. So when ONE file is converged across trees, the only sanctioned way to record it is to REGENERATE THE WHOLE MANIFEST -- and the record is explicit that a whole-tree regeneration while a sibling is mid-edit is the one action that blesses unreviewed bytes at scale. Measured the same day: nx_treediverge reports 2,154 divergent paths, so that precondition is not merely risky, it is UNMET. The result was a converged file that could not be recorded, and therefore an organ that could not be rebuilt: nx_sov_build_run, laptop and NAS byte-identical, still refused by the canon gate. ★A FAMILY WITH NO SURGICAL WRITE TURNS EVERY CORRECTION INTO A WHOLE-TREE REGENERATION, SO THE ONLY SAFE-SIZED FIX IS THE ONE NOBODY CAN MAKE. SCOPE, deliberately narrow: one row, named by its relpath, in a manifest whose other bytes are untouched. It does not scan a tree, does not compute a hash, and cannot invent a row. Recording a convergence someone else proved is a DIFFERENT act from deciding one, and only the first belongs here. license_tier: ORIGINAL Writes ONE named file, atomically. No hw writes (Rule 26). expect_exit: 0