Schema discovery
Schema inference is deterministic: the same input produces the same schema. Nested JSON, CSV headers, Parquet metadata, and mixed types are handled without manual mapping.
Evolution
When schemas change, Skippr diffs and applies backward-compatible updates. New columns are added. Type changes create a new column; existing data is not rewritten. Downstream queries keep working against the original column.
Pipeline
- Sample — read a configurable sample from the source.
- Infer — build a complete schema, including nested records, arrays, and maps.
- Diff — compare against the previous run.
- Apply — additive evolution only.
See How it works and Source landing semantics.