In live environments (like F1 racing or clinical monitoring), data producers often change types mid-stream (e.g., sending an int when the schema expects a float). Usually, this crashes the ingestion pipeline or forces a restart.
I built a "self-healing" layer that catches these errors, inspects the target schema, and auto-casts the data to a compatible type without dropping the record.
I just finished a stress test against live OpenF1 data. The results were:
-6 schema drift Events (unexpected type changes detected)
-0 system failures (all auto-recovered)
-100% data integrity
I also implemented a tamper-evident audit log (cryptographically signed) to prove that the "healed" data wasn't corrupted during the cast.
Repo (includes the new stress-test PDF report): https://github.com/tarek-clarke/resilient-rap-framework
The repo has already hit 1,500+ clones this week. I'd love to hear your feedback on the recovery logic!