Catch unexpected field changes in research exports — what should the operator record?
A renamed identifier column can break a downstream join even when the export contains all expected rows. Treat the column contract as a versioned operational dependency. Detect missing, added and retyped fields before a spreadsheet macro or analysis script silently produces a misleading but tidy-looking table.
Validate exported column names and meanings without silently discarding new or renamed fields.
Record column_name, logical_type, nullable and intended_use for each required output field; derive the names from actual returned data. Logical type should say opaque-string for identifiers and explain timestamp precision separately; a generic text label can hide assumptions needed by downstream comparison code.
If a legitimate rename is confirmed, update the local projection revision and record the change before releasing the downstream artifact. Preserve the old mapping reference for historical artifacts and state which outputs require regeneration, rather than making a global rename that obscures earlier interpretation.Write the contract
Record column_name, logical_type, nullable and intended_use for each required output field; derive the names from actual returned data. Logical type should say opaque-string for identifiers and explain timestamp precision separately; a generic text label can hide assumptions needed by downstream comparison code.
Inspect the header safely
Read headers with a CSV or JSON parser that preserves quoting and encoding rather than splitting a text line on every comma. A quoted header may contain punctuation, and a byte-order marker can affect the first field; inspect parsed names rather than trusting the first displayed line.
Classify contract drift
Distinguish missing required fields, unexpected additions and harmless ordering changes; retain unknown columns for review rather than discarding them without notice. A new field may carry sensitive material not appropriate for existing recipients, so schema additions need a disclosure review even when downstream code ignores them.
Test downstream behavior
Run the intended join on a permitted small working copy and verify that missing identifiers cause an explicit failure, not an all-null match. Check a fixture missing the key column and another with a legitimate empty optional column; only the former should necessarily block an identity-dependent join.
Version the accepted mapping
If a legitimate rename is confirmed, update the local projection revision and record the change before releasing the downstream artifact. Preserve the old mapping reference for historical artifacts and state which outputs require regeneration, rather than making a global rename that obscures earlier interpretation.
Keep the boundaries clear.
- Validate exported column names and meanings without silently discarding new or renamed fields.
- Operational record: Write the contract. Context badges do not certify access.
Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.
Primary references
Documentation reviewed 2026-11-15. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.