A JSON ingestion pipeline receives some records with missing required identifiers or invalid numeric fields. Valid records should continue through processing, but rejected records must remain available for correction and replay. Which approach is best?

1. Fail all future ingestion permanently after the first malformed record, even when valid records are independent.
2. Discard malformed records without logging or preserving their source.
3. Validate required fields and types, send invalid records to a quarantine path with reasons and source identity, and process valid records through the normal path.
4. Replace every missing identifier with the same constant and silently coerce all invalid amounts to zero.