nx_treediff.nx

nx_treediff.nx -- CROSS-TREE SOURCE DIVERGENCE CENSUS (laptop SSOT vs NAS buildroot). THE GAP IT CLOSES (filed seq1429, hit FOUR times in one session 2026-07-30): seq768 the build-admission enforcement was in NEITHER tree seq1393 five compiler defects existed ONLY in the buildroot copy seq1419 the deployed compiler was fine; the buildroot copy was the divergent one seq1429 nx_office_daemon/nx_survey_daemon were stale in buildroot while the SSOT was migrated nx_dup_source_check finds duplicate basenames WITHIN buildroot. NOTHING compared the buildroot against the laptop SSOT, so every one of those was found by hand, by accident, after the damage. WHY IT MATTERS BEYOND TIDINESS: /api/build compiles from BUILDROOT. A fix that lives only in the SSOT never reaches a running binary; a fix that lives only in buildroot is destroyed by the next sync from the SSOT. BOTH DIRECTIONS SILENTLY LOSE WORK, which is why divergence must be a REPORT, not an archaeology exercise. DESIGN: this half runs ON THE NAS and emits a manifest of `<bytes> <relpath>` for one tree, in a stable sorted-by-path order. The caller diffs two manifests (one per tree). Byte SIZE is a deliberately cheap discriminator -- it caught all four of today's instances -- and it needs no hash primitive on either side. A size match is NOT proof of identity; it is a screen that turns an unbounded search into a short candidate list, and it says so in its own output rather than implying certainty it does not have. FAIL-CLOSED + HONEST ENVELOPE: every cap is declared in-band (files scanned, cap hit, depth). A scanner that silently truncates its own coverage is the self-ceiling defect (law L011) -- so if a cap is reached the verdict SAYS SO and reports PARTIAL rather than presenting a floor as a total. DIALECT: plain-if, no empty literals, <=6 params, consts above use.