nx_wirecensus_lib.nx

nx_wirecensus_lib.nx -- S4 OF THE WIRING CENSUS: REGISTERED BUT NEVER CALLED. WHY THIS AND NOT MORE CALL-GRAPH (July-2026 external SOTA, debt id=1785451391): the field's largest dead-surface win came from RUNTIME evidence, not static analysis -- an Express codebase shed 16,000 lines (35 pct) by cross-referencing ACCESS LOGS against REGISTERED ROUTES, and the write-up states plainly that static tools "only detect unreachable code, not the absence of live traffic". nx_adopt already answers the static question (S1: defined, zero external callers). It structurally CANNOT see a tool that is registered, callable, advertised -- and that nobody has ever invoked. That tool is worse than dead code: it occupies the capability surface and reads as a feature. THE TWO WITNESSES, and why one is strong and one is weak (stated so the verdict is not oversold): W1 cap_consent.log "CAPMINT ... allow=<csv> ..." -- STRONG. A capability is MANDATORY to invoke a tool over /mcp, so a tool that never appears in any allow= list was never callable in practice. Absence here is close to proof. W2 actlog.jrnl TAB field 2 = the tool actually invoked -- WEAK BUT POSITIVE. Seats log VOLUNTARILY, so absence proves nothing; PRESENCE proves use. Used only to rescue tools from the dark list, never to condemn them. A tool is DARK only if BOTH witnesses are silent. Union, not intersection -- the conservative side. KNOWN BLIND SPOT, REPORTED NOT HIDDEN: caps minted before this log existed, or minted with the root nx_cap_mint on the NAS rather than POST /api/cap/mint, never reach cap_consent.log. So DARK is an UPPER BOUND on unused surface, exactly as nx_adopt's raw count was an upper bound before the public/local split. Anyone quoting this number must quote the bound with it. license_tier: ORIGINAL Read-only. No hw writes (Rule 26).