nx_paper_audit.nx

nx_paper_audit.nx -- RESOLVE every citation in a manuscript against its ledger (AS-3). ★WHY THIS EXISTS -- A MEASURED BLIND SPOT IN OUR OWN RULER, NOT A SUSPECTED ONE. nx_paperbench COUNTS provenance tokens; it never checks that a token RESOLVES, nor that the number printed next to it MATCHES the ledger row it names. Proven with a fixture (knowledge/papers/rt004_negcontrol_liar.md): a manuscript citing "999 [ev:E02]" when the ledger says 254, and "[ev:E99]" which does not exist at all, scored composite 750 and publishable=1. A well-formed token is indistinguishable from a true one to a counter. This organ closes that hole by a DIFFERENT METHOD: it resolves. For each "[ev:ID]" it 1. looks ID up in the ledger -> absent = DANGLING 2. reads the digit run immediately before -> differs = CONTRADICTED 3. no adjacent number -> UNANCHORED (reported, not an error: a citation may legitimately support a sentence rather than a figure) ★It audits the ARTIFACT, not the process. nx_paper_forge guarantees correctness at EMIT time; this checks the file as it now stands, so a hand-edited, post-processed, or forge-bugged manuscript is still caught. That independence is the whole point. DRY: ledger lookup is nx_paper_forge's pf_lookup and marker scanning is nx_paperbench's pb_find -- reused, not reimplemented (rule 15). Imports are transitive through the forge. Integer, deterministic. No hw writes (Rule 26). license_tier: ORIGINAL module: nishi-core.research.paper_audit depends: nx_paper_forge.nx (-> nx_paperbench.nx -> nx_syscalls.nx) genealogy_id: paperbench_token_count_blindspot + forge_ledger_lookup