`aip_sdk.TestPlanRecord`

aip_sdk.TestPlanRecord(id: str, workspace_id: str, name: str, version: int, plan: TestPlan, user_id: str, created_at: datetime | None, updated_at: datetime | None, applicable_quality_checks: list[str], required_columns: list[str], replayed: bool)

A version this call just published — what publish_test_plan and publish_test_plan_version return.

Carries what publishing can answer, and nothing a publish does not report. Reading a version back resolves it against the catalogue as it stands at that moment, which publishing does not do — get_test_plan returns a TestPlanDetail for that, and this type does not have those fields rather than carrying them unset.

Attributes

required_columns list[str]: Every column a dataset needs for the plan's metrics to score, as of this publish. Never stored: the platform derives it fresh from the metric catalogue on every read, so this same version can report a longer list once one of its metrics gains a required column. This value is therefore a snapshot of the moment you published; get_test_plan is the live read of the same version. An empty list means the plan genuinely needs no columns.
replayed bool: True when this is a prior publish's outcome, returned because this call's (internally generated) idempotency key matched an earlier one, rather than because this call stored anything new. In practice this only happens when this SDK itself retried a lost response — two ordinary publish_test_plan calls never share a key.