nx_regdup.nx

nx_regdup.nx -- DUPLICATE-KEY DETECTOR FOR ANY TSV REGISTRY. Fail-closed, additive, general. WHY. On 2026-08-07 the papers registry carried the identifier RT-012 TWICE, on two different manuscripts, and `nx_papers_index` rendered both sections without complaint -- so the live page published a collision as authority. That is this estate's own banked law: ★★★★★★ A GENERATED INDEX INHERITS ITS REGISTRY'S ERRORS AND LAUNDERS THEM AS AUTHORITY. The generator was not wrong; nothing was CHECKING. And the same session added a THIRD collision, because its author declared an id free after reading WINDOWS of the registry instead of enumerating it ⇒ ★★★★★★ A SAMPLE OF A FILE CANNOT ESTABLISH THAT A KEY IS FREE. ONLY AN ENUMERATION CAN. WHY A SEPARATE ORGAN AND NOT AN EDIT TO nx_papers_index. Measured the same day: the LOCAL and NAS copies of that organ have FORKED -- the local source emits a different page design than the one actually served. Editing the local copy and promoting it would have silently regressed the live page's design while "fixing" a duplicate. So this is additive: it adds an invariant nothing checked before, touches no serving organ, and can regress nothing by construction. ⚠It is NOT a second ruler for an existing invariant (that would be the duplicate-ruler defect) -- no organ in this estate checked registry key uniqueness at all. GENERAL ON PURPOSE. Every plane here is key-first TSV, and at least one other is known defective: the clockjobs plane is recorded at 67 rows over 50 distinct names, i.e. 17 surplus, where the LIVE row is the OLDEST because the registrar is idempotent-by-name. One ruler, every registry. nx_regdup check <registry.tsv> [keycol] exit 0 UNIQUE · 1 DUPLICATES · 3 unreadable nx_regdup selftest license_tier: ORIGINAL No hardware writes (Rule 26).