nx_asset_signals_gate.nx

nx_asset_signals_gate.nx -- KAT + TEETH for nx_asset_signals (R4 "what we need to do" pillar). Builds a FIXTURE catalog with KNOWN issues and proves the EXACT task signals are derived, that a fully-clean record yields ZERO signals (TEETH: no false positives), and that the derived signals can be EMITTED to a DEDICATED namespace and READ BACK (emission is verifiable end-to-end): record1 image, width+height present, has relations, is_current=1, HAS a phash, NO credential, tags EMPTY -> EXPECT exactly { MISSING_PROVENANCE, UNTAGGED } record2 image, width/height ABSENT, tagged, valid credential, has relations, is_current=1 -> EXPECT exactly { MISSING_METADATA } record3 image, full metadata, tagged, valid credential, has relations, is_current=1, phash = P record4 image, full metadata, tagged, valid credential, has relations, is_current=1, phash = P (record3 & record4 share an IDENTICAL perceptual fingerprint = a near-dup pair) -> EXPECT exactly { NEAR_DUP } for EACH of record3, record4 record5 image, full metadata, tagged, valid credential, has relations, is_current=1, UNIQUE phash -> EXPECT ZERO signals (the TEETH: a clean record triggers nothing) Assertions: (a) every EXPECTED (cid, code) signal is present in as_scan's output (b) NO UNEXPECTED signal exists (exact-set: total derived count == expected count) (c) record5 contributes ZERO signals (clean-record teeth, checked explicitly by cid) (d) as_emit writes them into the DEDICATED asset-signals namespace and as_list reads them ALL back, and as_get on one key returns its TAB record with the right signal_code+cid fields (verifiable) (e) ISOLATION: the production board store (knowledge/store/ws-) is NEVER touched (the gate uses a fresh /tmp catalog AND the dedicated namespace prefix is a different store by construction). Verdict appended to knowledge/status/asset_signals_gate.log (additive law #13). expect_exit: 0 license_tier: ORIGINAL