Rebuild input datasets of recipes whose output schema may depend on input data (prepare, pivot, …)

Rebuild output datasets of recipes whose output schema is computed at runtime (code, pivot, …)

Always perform expensive checks