JSONL to CSV with late fields kept.

The difficult example: a field appears after a blank line

Blank lines should not become records, and a field introduced later must still get a column. The union schema below discovers payload.amount and trace even though neither exists on the first record.

{"id":1,"payload":{"kind":"pageview"}}

{"id":2,"payload":{"kind":"purchase","amount":19.95},"trace":"t-9"}

id,payload.kind,payload.amount,trace
1,pageview,,
2,purchase,19.95,t-9

Preview caveat: when the browser preview samples a large stream, it is labeled provisional. The complete conversion uses the field option you selected and reports malformed input instead of silently skipping a line.

Operational details that matter

• Union across records: late keys are columns, with empty cells only where a record truly has no value.
• Numbers stay exact: identifiers and exponent notation remain as written.
• Malformed input: invalid JSON is reported instead of silently skipped.