validate-data

Export the pipeline schema as a Mermaid diagram with dlthub local pipeline schema, then review tables and sample rows in the Workspace Dashboard or query them through the pipeline MCP server. Fix issues with processing_steps and re-run the pipeline.

Use processing_steps in the resource config with map, filter, or yield_map to transform items before loading. For monetary or precision-sensitive values, always convert to Decimal rather than float to avoid rounding errors.

dlt auto-unnests nested arrays into child tables named like <resource>__results, which works well for analytics. If you need a flat structure, use yield_map to flatten items or adjust data_selector to point deeper into the response.

Columns that are entirely null on the first load get no inferred type and may be absent. Add columns hints to the resource config with an explicit data_type, or adjust API parameters like group_by so the fields are populated.

Enable dev_mode so each run produces a fresh dataset, then re-run the pipeline and inspect the result with the Mermaid schema export or MCP queries. Use the debug-pipeline skill to examine traces and load packages between iterations.