nx_discocensus.nx

nx_discocensus.nx -- DISCOVERABILITY CENSUS (2026-08-06). Measures, on every run, how much of the tool registry is actually USABLE by an agent: does each tool carry a call contract, and is its safety profile derived rather than fabricated? WHY THIS ORGAN EXISTS -- the failure it is built to prevent, with a receipt. Debt 1784549913 filed the missing-col8 gap when the estate had 111 MCP tools and 21 typed contracts. Nobody re-measured. By 2026-08-06 it was 749 of 869 -- the SAME defect, EIGHT TIMES LARGER, sitting open the whole time. ★★★★★★ AN OPEN DEBT WHOSE MEASURED SIZE GREW 8x WHILE NOBODY RE-MEASURED IT *IS* THE DRIVE-THE-PLAN FAILURE. A number that is only ever measured by hand is measured once. This makes it a standing instrument so the PM sees the trend without a human deciding to look. Siblings, deliberately NOT duplicated: nx_adopt answers "which primitives has nobody adopted", nx_wirecensus answers "registered but never called". This answers "discoverable ENOUGH TO CALL". HONEST BY CONSTRUCTION: every figure is a COUNT of rows that exist on disk -- nothing is asserted, and a missing input file is REFUSED, never silently treated as zero coverage (a zero you cannot distinguish from an unreadable file is the bounded-read defect all over again). Floors live in a CONF, not in code (rule 11); with no conf the verdict is UNKNOWN and the census still prints -- it will not invent a bar to grade itself against. nx_discocensus -> JSON census on stdout; exit 0 GREEN/UNKNOWN, 1 RED (a floor was breached) license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0