nx_capreach.nx

nx_capreach.nx -- WHEN A TOOL IS DENIED, WHAT ELSE REACHES ITS FUNCTION? (2026-09-04) THE DEFECT THIS CLOSES, measured FIVE TIMES IN ONE SESSION BY ONE SEAT. A capability denial names a TOOL. It does not name the FUNCTION that tool performs, and in this estate the function is almost always reachable another way -- but a denial message ends the search, because it reads like a wall and offers only "mint a cap". Measured instances, all 2026-09-04: nx_capsearch denied, reachable because nx_spendgate FORKS it; nx_sov_build_run denied, reachable because nx_stale_check forks it; the deploy lane denied, reachable through nx_restage; nx_sota_status denied, its VERDICT readable because the organ runs on a beat and PERSISTS it to knowledge/status/sota_board.ledger. Each time a blocker was published before the reach was enumerated, and each time the reach existed. A FALSE I-AM-BLOCKED IS SELF-CONFIRMING, because nothing afterwards tests it. TWO RELATIONS, AND THEY ARE THE TWO THAT ACTUALLY WORKED -- not the one that was easiest to write. A SHARED-ELF detector was considered and REJECTED: tool_allowlist.conf does map several names onto one elf (nx_status and nx_torstat both point at nx_hostctl), so it is trivially implementable, but NONE of the five measured cases had that shape. Shipping it would have been a detector that cannot fire on the cases that motivated it -- a different detector wearing the name of the problem. ARTIFACT the target's own persisted output, which carries its verdict when the organ cannot be called. *** SHIPPED IN THIS LEG. *** FORK a source that names the target and is itself a registered tool: a callable composer. *** NOT IMPLEMENTED IN THIS LEG, AND SAID SO HERE RATHER THAN LEFT IMPLIED. *** It needs a whole-corpus scan and therefore its own cost envelope, so it is a NAMED NEXT LEG. This paragraph exists because the first draft described BOTH relations as if both shipped -- the exact overclaiming-prose defect that nx_board_contract_gate's prose axis was built THE SAME DAY to catch. A header is not a measurement, and an organ whose own comment oversells it is the cheapest possible place for that defect to start. PUBLISHES A LIST, NEVER A VERDICT. The estate has measured that a ranker whose top hit is not the answer must print evidence and let a human decide (nx_spendgate ships with no threshold for exactly this reason). A composer that merely NAMES the target may not expose it; this organ says so rather than scoring it. exit: 0 candidates printed . 2 usage . 3 no allowlist readable license_tier: ORIGINAL No hw writes.