nx_gfxpapers_ingest.nx

nx_gfxpapers_ingest.nx -- MIRROR THE GRAPHICS PAPER INDEX INTO THE ESTATE LIBRARY SO IT CANNOT ROT. Operator 2026-08-15: ingest kesen.realtimerendering.com "in a non rottable way", and then -- correctly -- "shouldnt these be in papers and our library?". The first cut wrote to knowledge/gfxmirror/, an ad-hoc directory beside the three homes the estate already has, which is the orphan the no-orphans doctrine forbids and which nx_spendgate had already warned about by naming nx_papers_index as the top incumbent. The homes, and why THIS content belongs in the middle one: knowledge/fetched/ -- raw bodies from nx_research_fetch, transient working material knowledge/library/ -- CURATED external reference documents, topic-prefixed, .fail markers for misses knowledge/papers/ -- .nxpap native papers (nx_paper_native) plus authored .conf/.md Ke-Sen Huang's index is the canonical catalogue of real-time-rendering conference papers (SIGGRAPH, SIGGRAPH Asia, EG, I3D, EGSR, SCA, SGP, HPG, PG, NPAR, SMI, GI; 2000-2026). Its venue pages are reference documents, so they land in the library under a caller-supplied prefix that keeps them attributable in a flat namespace shared with 7,000 other references. WHAT "NON-ROTTABLE" MEANS, precisely: a bookmark is not an ingest. A URL list rots the moment a site moves or a domain lapses, and the failure is SILENT because a dead link and an unvisited link look identical in a registry. This stores the BYTES with their provenance. MEASURED on the first run: one indexed page (conference2012.html) was ALREADY 404 on capture day, and the entire www.realtimerendering.com host returns 403 to this client for all 236 pages while kesen.realtimerendering.com serves them -- so a registry of the hrefs exactly as the index writes them would have been 236 dead entries that look healthy. CLOSURE, NOT ONE HOP. "All of its content" is a fixed point, not a depth guess: every mirrored page is itself scanned for further pages in the SAME namespace and the queue grows until a pass adds nothing. The anchor is the site's own `kesen/` path segment, so the crawl can never wander onto third-party hosts -- mirroring every paper PDF this index points at would be a different, and far ruder, program. WHAT IT DELIBERATELY DOES NOT DO: parse paper titles. A brittle HTML scrape would be an unvalidated measurement dressed as data. Bytes and provenance are exact and checkable today; extraction is a later rung that can run against the mirror OFFLINE, repeatedly, without re-fetching a single page. argv: <index-url> <out-dir> [fetcher-elf] [name-prefix] license_tier: ORIGINAL No hw writes (Rule 26).