nx_licdetect_lib.nx

nx_licdetect_lib.nx -- WHICH LICENCE IS THIS TEXT? Read from the bytes, never from a declaration. THE MEASUREMENT THAT JUSTIFIES A TEXT DETECTOR AT ALL. Across 124,278 licence chains surveyed in the field, 96.5% of datasets and 95.8% of models LACK the required licence text, and only 2.3% / 3.2% satisfy both the licence-text and copyright requirements. The survey's conclusion is this lib's design rule: LICENCE FILES AND NOTICES, NOT METADATA, ARE THE SOURCE OF LEGAL TRUTH. It is also what the librarian measured directly: of 26 NAVER releases, 2 carried any licence string on the listing, and the one we then FETCHED turned out to be 483 bytes off the canonical Apache text because it adds a copyright preamble and drops the appendix. A declaration is a claim; bytes are not. THREE OUTCOMES, AND ONLY ONE OF THEM IS AN ANSWER: DETECTED exactly one licence's REQUIRE markers all matched and none of its EXCLUDE markers did UNKNOWN nothing matched -- fail-closed, the artifact does NOT advance to evidence=READ AMBIGUOUS two or more matched -- ALSO fail-closed. A text matching two licences is not a decision, and taking the first would be a coin flip wearing a verdict. MARKERS ARE DATA (knowledge/license_markers.conf), so a new licence costs a row and never a rebuild, and every marker set is auditable by a human reading the conf rather than the binary. NO GUESSED CAPS: every array here is sized from the CONF'S OWN LINE COUNT, counted before allocation. A marker table cannot overflow a bound derived from the file that defines it. license_tier: ORIGINAL