Erasure
Remove one classifier's data, bundles and complete artifacts, including interrupted external runs.
Request erasure
await classifier.erase();Erasure is scoped to the classifier name. Other classifiers sharing the data directory remain intact.
The operation:
- prevents new work from joining the old data generation;
- records erasure intent durably;
- cancels an active provider run and verifies cleanup;
- verifies remote resource cleanup;
- removes examples, protected tests and run records;
- removes raw runner bundles and every file in candidate and promoted artifacts;
- creates a fresh empty data generation.
It does not report completion while an external run can still write an old artifact. If provider cancellation or cleanup cannot be verified, erasure remains pending and exposes the failure for retry.
Filesystem boundaries
SwapAI deletes only paths recorded for that classifier and verifies they remain inside its managed data directory. It does not follow arbitrary paths from a runner manifest.
Operational evidence
Inspection exposes erasure state and any cleanup failure without exposing raw data. Treat completion as the point at which the method resolves successfully, not the time at which a request was first queued.