nx_absent.nx

nx_absent.nx -- PROVE ABSENCE, OR REFUSE TO CLAIM IT. Sovereign, no shell, no client-side filtering. WHY THIS EXISTS, measured on myself 2026-08-07: three times in one day I ran a search, piped it through a filter, saw no hits, and concluded the thing was ABSENT. Every time the tool had HONESTLY declared its own partiality -- `NX-TRUNCATED OUTPUT-IS-PARTIAL capture_cap=163840`, `BUDGET-EXCEEDED partial=1`, `coverage_complete=0` -- and every time my filter discarded that line before I read it. Once it made a grep over 22,000 files read as a clean absence and I began doubting a correct census; once I was one step from filing a sev-9 that was flatly untrue. THE SEARCHER WAS NEVER THE DEFECT. THE READER WAS. So this organ does not re-implement search: it COMPOSES nx_shelltool (already honest, already coverage-declaring) and moves the verdict OUT OF THE PAYLOAD AND INTO THE EXIT CODE. ★★★★★★ AN ADVISORY IN A PAYLOAD IS DISCARDABLE; AN EXIT CODE IS BRANCHED ON. If a fact must not be ignored, it cannot live only in text that a caller is free to grep away. ★★★★★ `matches=0` IS NOT ABSENCE -- IT IS ABSENCE *WITHIN WHATEVER WAS SCANNED*. Absence is a claim about the WHOLE corpus, so it requires proof of complete coverage, and this organ will not issue that claim without it. THREE STATES, because two would force a lie (the ecosystem's own law: a probe that cannot tell MET from UNOBSERVABLE reports failure as success): exit 0 ABSENT-PROVEN 0 matches AND coverage_complete=1 AND no truncation marker exit 1 PRESENT >=1 match (the corpus contains it; absence is refuted) exit 3 UNPROVEN coverage was partial -> NO CONCLUSION IS AVAILABLE, in either direction exit 2 usage UNPROVEN is deliberately NOT folded into either answer. Folding it into PRESENT would cry wolf; folding it into ABSENT is exactly the mistake this organ was built to make impossible. usage: nx_absent <pattern> <dir> [ext] license_tier: ORIGINAL Read-only: forks a read-only searcher, writes nothing. (Rule 26)