nx_4chan.nx

nx_4chan.nx -- SOVEREIGN 4chan BROWSE / PREVIEW / DOWNLOAD adapter: the first of the media-manager per-source ingesters ported off Python (mm/src/sites/fourchan.py) into the estate. Emits JSON on stdout so the gallery/vault surface consumes it directly; downloads land in a plain directory that `nx_mvault_walk <dir> real 4chan commit` then records into the vault (CID dedup + DSM layout). COMPOSITION, NOT REIMPLEMENTATION: the vault already owns classification/dedup/layout (B1-B10), so this organ deliberately stops at "validated bytes in a directory" and hands off. nx_4chan boards -- board list (+nsfw flag) nx_4chan catalog <board> [max] -- thread previews: subject, comment, counts, THUMB+FULL urls nx_4chan thread <board> <threadno> [max] -- every media item in a thread nx_4chan fetch <board> <threadno> <dir> [max] -- download the thread's media into <dir> nx_4chan selftest -- offline gate: parser teeth incl. neg-controls ★ WHY A BRACE-DEPTH PARSER AND NOT KEY-SCANNING. catalog.json embeds a `last_replies` array whose post objects each carry their OWN "no": field, so scanning for `"no":` reports replies AS THREADS -- and it fails silently, inflating a 15-thread page into 60 "threads". Depth tracking is immune to field order and to nesting. It is also STRING-AWARE: a 4chan comment routinely contains { } and ", and a structural scanner that reads braces inside string data mis-nests the whole document. Conveniently ALL THREE endpoints put their objects of interest at brace-depth 2: catalog.json [ {page {1}, threads:[ {thread {2}, last_replies:[ {reply {3}} ]} ]} ] thread.json { {1} "posts":[ {post {2}} ] } boards.json { {1} "boards":[ {board {2}} ] } so one scanner serves all three and `want_depth` stays a parameter rather than three constants. LEGAL/RATE POSTURE: read-only public JSON API, one polite delay between media fetches. General capability; the operator points it where they like. license_tier: ORIGINAL