nx_docprose.nx

nx_docprose.nx -- LIB: WHERE DOES A STORED DOCUMENT'S REAL CONTENT BEGIN? THE DEFECT THIS EXISTS TO REMOVE, MEASURED LIVE ON nishifamily.com/search 2026-08-25 (site scope, one result page): results were rendered with these as their TITLES -- "HTTP/1.1 200 OK Date: Thu, 02 Jul 2026 17:07:07 GMT Content-Type: text/h" "<p align=\"center\">" The renderer was not at fault: it faithfully took "the first non-empty LINE", and the first line of those stored documents really IS an HTTP status line / a raw markup tag, because the evidence-mirror captures in knowledge/fetched are indexed as documents with their wire bytes intact. ROOT vs REACH -- STATED PLAINLY SO NOBODY MISREADS THIS LIB'S SCOPE: The ROOT fix is at INGEST (derive clean text before indexing; nx_block_density's bd_fit_text is the estate's shipping boilerplate remover and is the organ for that job). This lib is the SERVE-SIDE half, and it is NOT a workaround: it repairs the ALREADY-INDEXED corpus with no reindex, and it stays correct afterwards as defence in depth. An ingest fix alone leaves every existing document broken until a full reindex; a serve fix alone leaves the stored text dirty for every other consumer. Both are owed and both are real. POLICY IS CONF, STRUCTURE IS CODE. The protocol facts (an HTTP header block ends at the first blank line, RFC 9112 2.1) are code. The one judgement call -- how many letters make a "word" -- is a conf row. license_tier: ORIGINAL