tsv.nx

tsv.nx -- Tab-separated values reader + writer. The "database format" for sovereign apps that don't yet have a SQLite binding (per ROADMAP P8). Append-only files, one per entity (people.tsv, obligations.tsv, interactions.tsv). Reads are linear scans; writes are fwrite-append at end. Format (RFC 4180 TSV variant, simpler than CSV): - Fields separated by TAB (0x09) - Rows separated by LF (0x0A) - No field-level quoting -- TAB and LF inside fields escaped via \t and \n (our convention; not RFC 4180) - Leading #-comment rows skipped by the reader Not designed for megabyte-scale; linear scans are O(n). Fine at the ~1k-record scale that CLAUDE.md rule #1 (Schema First) implies for a CRM/ops app's first iteration. Invariants: TSV1 Round-trip exact for any row whose fields don't contain raw TAB or LF. Fields with TAB/LF are lossy unless caller pre-escapes via tsv_escape_field. TSV2 Comment rows (# at start-of-line) never contribute fields to row callbacks. TSV3 Trailing newline on the file is tolerated; absence is also fine. TSV4 Reader is streaming: processes byte-by-byte without buffering whole rows. Safe on arbitrarily-long files.