nx_srcreg_lib.nx

nx_srcreg_lib.nx -- THE ONE READER FOR THE RESEARCH SOURCE REGISTRY (knowledge/research_sources.conf). Operator 2026-09-06: "we need to measure and make sure we never have any unreachable data whether by search or api we need to be able to gather it all" and "a regular cadence for all domains without smashing our nas resources". The registry is the POPULATION every reach and coverage number on /compare/researchreach is measured over, so a reader that silently drops a row it cannot parse under-reports coverage in the FLATTERING direction. This lib refuses a bad row BY NAME instead, and its partition must sum to the rows seen. THE DATA (rows in knowledge/research_sources.conf, pipe-separated, NO pipe inside a field): src|<key>|<kind>|<protocol>|<base-url>|<pace_ms>|<auth>|<licence>|<note> kind forge | preprint | index | publisher | government | repository | registry protocol oai-pmh | atom-api | rest-json | graphql | sitemap | html pace_ms MINIMUM milliseconds between two requests to THIS host. Per-source and not global, because a shared budget starves the slow hosts first. THE STATES, per row: OK nine fields, a declared kind, a declared protocol, a positive integer pace BAD-FIELDS fewer than nine fields -- the row cannot be read, and is NOT silently skipped BAD-KIND a kind outside the declared vocabulary BAD-PROTO a protocol outside the declared vocabulary BAD-PACE a pace that is not all digits, or is zero or negative A row belongs to exactly ONE bucket and rows = ok + bad_fields + bad_kind + bad_proto + bad_pace must SUM. The vocabularies are CLOSED on purpose: an unknown protocol must REFUSE rather than default to the permissive value, because a mistyped protocol that falls through to a default is a silent downgrade of a fail-closed row. COMPOSES, never re-implements: sys_read_file (sizes its buffer from the file and cannot short-read, so there is no cap to reach in silence here). license_tier: ORIGINAL No hw writes (Rule 26).