Research Data Intake Validation Agent

Research datasets often arrive with schema, unit or provenance problems that are only discovered months later, after analysis has started. This agent runs when a dataset is received. It validates the file structure and metadata against the data dictionary, then runs unit and range checks on every column without touching the raw files. Missing units, unknown sources or impossible values are logged as exceptions for the researcher, who decides what they mean and whether anything should be excluded. Only transformations the researcher has approved are applied, and only to a cleaned working copy, which is checked again after each transformation. Edge case: a temperature column with no unit is logged, never assumed to be Celsius because the numbers look plausible.