nx_aritycheck_gate.nx

nx_aritycheck_gate.nx -- THE ARITY CENSUS: report every CALL SITE whose argument count disagrees with the definition its own directory resolves. The axis nothing currently measures. WHY THIS EXISTS (measured 2026-07-31, lib-reconcile). nx_cc is fails-open on arity: seq1012 shows undefined identifiers compile clean and 1785447657 shows duplicate definitions compile clean with zero diagnostics. nx_undefscan closes the "called but never DEFINED" axis and is the right tool for it -- but it reports VERDICT=CLEAN on nx_input_abstract_gate, which passes ONE argument to a TWO-parameter ia_init at eight sites, because ia_init IS defined in the closure. The target also compiles. So nx_cc says fine, undefscan says CLEAN, the artifact exists, and the call is still wrong. ★★★★★CLEAN + BUILDS != CORRECT WHEN NO INSTRUMENT MEASURES THE AXIS THE DEFECT LIVES ON. SCOPE, stated so a caller never has to guess: ONE DIRECTORY. An import binds the importer's OWN directory first (proven from compiler output at buildroot/_build/nx_f32_llm_serve.s:2511), so the definitions visible to a file are, to first order, the ones beside it. That makes a per-directory census sound for the common case and DELIBERATELY BLIND to cross-layer resolution -- which is reported as UNKNOWN-NAME rather than silently ignored, because a name this organ cannot see is not a name it may bless. KNOWN FALSE POSITIVES, bounded and declared (the nx_undefscan doctrine, inherited on purpose): - a call through a FUNCTION-POINTER VARIABLE looks like a call to a name it cannot resolve - a name that is BOTH a local variable and a function elsewhere in the dir Both are bounded by the POSITIVE CONTROL: nx_m2d_engine calls ia_init(ia, 5) correctly in the SAME directory against the SAME definition and MUST report zero. A checker that flagged everything fails it. ⚠STATUS 2026-07-31: DESK-CHECKED, NEVER COMPILED. Authored while /mcp was wedged, so it has not been through nx_cc or its own selftest even once. Do NOT treat "reviewed" as "works" -- that is the exact confusion this organ exists to remove. WHAT THE SELFTEST MUST PRINT on a first good build: defs collected >= 3 (two_p=2, bad_caller=0, inner=2, good_caller=0) EXACTLY 1 arity mismatch: bad.nx two_p called with 1 arg(s), defined with 2 0 conflicting duplicates in /tmp/nxarity, and EXACTLY 1 in /tmp/nxarity2 If it reports 2 mismatches, ac_args lost its depth-awareness and counted the comma inside inner(1, 9); if it reports 0, ac_find or ac_is_def_at is failing to see the definition at all. Both failure modes are distinguishable from the message alone, on purpose. FIRST REAL TARGET once built: buildroot/runtime/_hdl_build -- it must reproduce the 8 ia_init mismatches in nx_input_abstract_gate (pre-move) and report ZERO for nx_m2d_engine. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0 when clean