nx_coedit.nx

nx_coedit.nx -- COLLABORATIVE-EDITING KERNEL (office census OF-C1's engine half): an RGA text CRDT. Multiple replicas edit the SAME document concurrently; their op logs MERGE deterministically -- the same final text on every replica REGARDLESS of merge order (convergence), no lost updates, tombstoned deletes, deterministic (counter,replica) tie-break for concurrent inserts at the same point. OP LOG (append-only, LF lines; the natural fit for the additive-only store doctrine): I <replica> <counter> <afterid> <char> insert ONE byte after node <afterid> ("R" = document root) D <replica> <counter> <targetid> tombstone node <targetid> ids are "<replica>:<counter>"; an op's id must be UNIQUE (same id + different content = LOUD protocol violation; identical duplicate = idempotent skip, which is what makes re-merging safe). <char> is one byte, escaped: \n=newline \t=tab \s=space \\=backslash (any other 1 byte literal). COMMANDS apply <log> <out.txt> materialize (RGA linearization) -> file + verdict line ins <log> <rep> <ctr> <afterid> <char> append a raw insert op del <log> <rep> <ctr> <targetid> append a raw delete op merge <A> <B> <out> union by op id (idempotent dedup; conflicting id = LOUD) edit <log> <rep> <pos> ins <text> position-based convenience: per-BYTE ops chained after the edit <log> <rep> <pos> del <count> visible byte at pos-1 / tombstones for [pos,pos+count) HONEST SCOPE (v1, named): the KERNEL -- byte-granular ops (UTF-8 sequences survive as consecutive ops but byte-position edits must not split a sequence), logs are plain files, ops cap 65536/doc. LIVE multi-client sync (the R2 daemon + browser surface) is NOT built yet; the census OF-C1 axis stays ABSENT until R2 -- its needle requires the R2 sync function DEFINITION, which this file intentionally does not contain. license_tier: ORIGINAL