nx_treediverge.nx

nx_treediverge.nx -- THE MISSING CROSS-TREE DIVERGENCE DETECTOR (debt 1785438829, sev-8). THE GAP IT CLOSES, in that debt's own words: "nx_dup_source_check finds dups WITHIN buildroot but there is NO laptop-vs-NAS divergence detector -- I did it BY HAND four times today. BUILD THAT: a cross-tree census emitting per-file laptop_bytes / nas_bytes / delta / direction, so divergence is a REPORT rather than an archaeology exercise; size alone would have caught all four of today's instances." WHY IT MATTERS (not tidiness): /api/build compiles from buildroot, so a fix living only in the laptop SSOT NEVER REACHES A RUNNING BINARY, and a fix living only in buildroot is DESTROYED by the next sync from the SSOT. The divergence is BIDIRECTIONAL -- neither tree is a superset -- so a one-way sync in EITHER direction silently loses work. You cannot merge safely without first knowing the direction per file. That is exactly what this emits. COMPOSES, DOES NOT DUPLICATE: nx_treediff already emits one tree's manifest as "<bytes> <relpath>" per .nx file and says in its own usage that it is "for cross-tree comparison" -- it just never performed the comparison. This is the missing half. Run nx_treediff on each tree, then: nx_treediverge <manifestA> <manifestB> [labelA] [labelB] VERDICTS, per file: A-ONLY / B-ONLY / A-AHEAD / B-AHEAD / SAME. "AHEAD" is BY BYTES, which is the signal that caught all four historical instances; it is a FLOOR, not proof of content (equal size can still differ -- see the same-size-rewrite caveat on expect=<size> CAS). Declared, never implied. NON-VACUITY: an empty or unreadable manifest REFUSES (exit 3) instead of reporting "0 divergent", because a comparator that reads nothing and prints GREEN is the self-ceiling defect this ecosystem has banned repeatedly. A zero-divergence verdict is only emitted when BOTH sides actually had rows.