@plurnk/plurnk-mimetypes-application-jsonl

application/jsonl (JSON Lines / NDJSON) mimetype handler for the plurnk ecosystem. Hand-rolled, no parser dependency.

install

npm i @plurnk/plurnk-mimetypes-application-jsonl

what it does

One JSON value per line — training data, eval sets, fine-tune files, chat/agent logs. The structural definition of a JSONL dataset is its record schema, not its rows (a file can be millions of lines), so:

• extractRaw(content) — the schema: each distinct top-level key across records becomes a field symbol at the line it first appears. Scale-safe (bounded by schema width, not record count), and it answers "what is this dataset" — {prompt, completion, score}.
• extent — the record count, the unit you address by.
• deepJson(content) — the parsed records array, a jsonpath target ($[5].completion), computed only on demand.
• query(content, dialect, pattern) — jsonpath dispatches against the records array; regex/glob against the raw text.

Lenient by design: blank lines are skipped, and a line that doesn't parse is skipped (a trailing newline or a partial write doesn't poison the file). The raw body is already readable JSON-per-line, so there is no content projection. References are not applicable.

license

MIT.