nx_debtlive.nx

nx_debtlive.nx -- IS THIS DEBT ALREADY FIXED? A mechanical closure check for the debt ledger. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0 THE DEFECT THIS EXISTS TO KILL: the debt ledger is the first thing a seat reads and the last thing that gets corrected. Fixes land; the rows that described them stay `open`. MEASURED 2026-07-31 (ws=mgmt-staging-cas): of the FIVE top sev-9 rows picked off `nx_debt sev 9`, FIVE were already fixed in source and FOUR were verified LIVE in the deployed binaries by string-grep of the artifacts themselves -- 1785453352 unpack staging CAS -> live in nx_mgmt_api.elf (pins the unpack to YOUR source) 1785450506 deploy takes no lease -> live in nx_mgmt_api.elf (acquire mgmt-api-deploy 600) 1785447965 rollback ignores target -> live in nx_mgmt_api.elf (ONE reverse gear) 1785437278 unpack must not backdate -> live in nx_treepack.elf (REFUSED-WOULD-DROP-SYMBOLS) 1785451053 4 MiB piece truncation -> fixed in source, PCAP 4 MiB -> 32 MiB plus a drop COUNTER Every one cost a fresh session the same rediscovery: read the row, believe it, re-derive the fix, then find it already shipped. That is not bookkeeping -- it is the ledger ACTIVELY MISDIRECTING the scarcest resource in the ecosystem, from the boot path, to every seat, every session. LAW: A LEDGER THAT CANNOT CLOSE ITS OWN ROWS BECOMES A LIST OF SOLVED PROBLEMS. WHY THIS IS DECIDABLE (the join key, and it is exact): fixers annotate the fix SITE with the debt row index -- seq1467, seq1379 HALF-2, seq1734, seq1807 CAS. And nx_debt row index IS that number: `nx_debt eat 1785451053` answers at_idx=1734 while nx_torrent_get.nx carries seq1734 on the line that fixes it. Verified on three independent pairs before a line of this was written. So the set of seqN tags in the tree is the set of debts the tree CLAIMS to have fixed; intersect with rows still `open` and you have the stale-ledger set. WHAT IT DOES NOT DO, ON PURPOSE: it NEVER eats a debt. A seqN in source proves a fixer CITED that row -- not that the fix is correct, complete, or deployed. Auto-eating on a comment would swap one lying instrument for another, and a false EATEN is worse than an open row. The verdict is CANDIDATE-CLOSED: a queue for an agent to confirm by ARTIFACT and then eat deliberately. Corroborates; never convicts. NON-VACUITY (this instrument can fail, and says so): a scanner that matches nothing reports no-stale-rows forever while the ledger rots -- the redseen=0 / coverage-gaming / silent-512KB-cap class. So: zero seq tags found -> INSTRUMENT-BLIND exit 2, no verdict; zero rows loaded -> INSTRUMENT-BLIND exit 2; declared rows > found rows -> coverage_complete=0 and verdict PARTIAL, never a clean bill. nx_debtlive scan [prefix] [dir1] [dir2] full sweep -> JSON envelope plus bounded candidate list nx_debtlive check <idx> [dir1] [dir2] one row: does the tree cite seq<idx>? exit 0=cited 1=not