Write semantic contracts
A schema should express meaning as well as storage type. A floating-point age field can still contain negative values or a newly changed unit. Record required fields, ranges, missing-value rules, category policy, and who owns each feature. An inferred schema describes observed data; review it before treating it as the intended contract.
Define checks for an order-total feature, including currency and whether refunds can be negative.
Separate blocking failures from signals
A missing identifier may invalidate every join and justify quarantine. A moderate shift in category frequency may warrant investigation while processing continues. Decide severity and ownership before the incident. Avoid silently dropping invalid records: losing one population can create a biased training sample.
A batch has 12% missing customer IDs concentrated in one region. Explain why simply dropping those rows is risky.
Validate both sides of transformations
Check raw inputs and transformed features. A valid raw category may map to an unexpected unknown token, or a unit conversion may produce extreme scaled values. Keep a small set of known examples with expected outputs. Compare distributions, but also verify joins and record counts because distribution checks alone can miss duplication.
A join doubles the row count without changing each numeric distribution. Name a check that catches it.
Make recovery explicit
Retain the failing batch identity, validation report, upstream version, and last accepted data snapshot. Repair the producer or transformation, then rerun validation before releasing the batch. Do not relax a contract merely to make a pipeline green; record why an intentional schema migration is compatible with downstream consumers.
Draft a recovery checklist for a cents-to-dollars producer change. Include a consumer parity check.
An inferred schema accepts a negative age. What should happen next?
Keep exploring
Use the official certification page and its linked resources to validate your coverage. For guided labs, explore the Google Skills learning path. Revisit the practical exercise and explain your decision in your own words before moving on.