nx_gfxpapers_extract.nx

nx_gfxpapers_extract.nx -- TURN THE MIRRORED GRAPHICS LISTINGS INTO A CITABLE PAPER INDEX. The library mirror (nx_gfxpapers_ingest) stores BYTES, which is what makes the corpus non-rottable but does not yet make it usable: /compare needs to cite a specific paper, not a 360 KB page. This organ reads the mirror OFFLINE -- no network, no re-fetch, runnable as many times as it takes to get the parse right, which is precisely why the ingest deliberately stored bytes and deferred extraction rather than doing a brittle scrape on the wire where every correction costs another crawl of someone else's server. THE ANCHOR IS `<dt><B>` ... `</B>`, AND IT WAS CHECKED FOR VARIANCE BEFORE IT WAS TRUSTED. Verified by reading real mirrored bytes at both ends of the corpus -- i3d2005Papers.htm (2005, .htm) and sig2025.html (2025, .html) -- so it is not a pattern fitted to the one page that motivated it. Both cases are counted and reported SEPARATELY rather than merged, because a case split that is silently accepted would halve a corpus without ever showing up as an error. ⚠HTML COMMENTS ARE SKIPPED, AND THAT IS NOT OPTIONAL. Every one of these pages carries a commented-out TEMPLATE entry near the top with the same tag shape as a real one. A scanner that does not skip comments measures the documentation, not the code -- the estate has paid for this law twice, and this session paid for it a third time when a contract symbol written in a COMMENT flipped a published compare row to landed with no capability behind it. ⚠TITLES ARE STRIPPED OF TAB AND NEWLINE. The row format is tab-delimited, so a title containing the delimiter would silently split into phantom columns -- a producer must never emit the byte its consumer uses as a separator, which is the same producer/consumer wire defect that has bitten the status planes. argv: <library-dir> <prefix> <out-index> license_tier: ORIGINAL No hw writes (Rule 26).