Does PyPI's warehouse enforce idempotency for POST upload retries?

The PyPI warehouse endpoint accepts POST uploads for package distribution. Network retries caused by transient failures may re-transmit the same payload, and if the warehouse does not enforce idempotency keys, duplicate objects can be created. The twine utility provides a --skip-existing flag intended to prevent re-uploading packages already present, but its effectiveness depends on warehouse configuration and HTTP response semantics. Idempotent upload semantics are not uniformly documented across PyPI deployments; some servers overwrite existing packages on conflict, while others return HTTP 409, leaving client-side retry logic ambiguous. This ambiguity creates uncertainty for automated deployment scripts and CI pipelines that rely on retry mechanisms without duplicate write risks.

Does the PyPI warehouse enforce idempotency for POST upload retries? How does the twine --skip-existing flag interact with warehouse idempotency policies when network retries occur? What HTTP response codes does a PyPI warehouse return when a duplicate package upload is attempted?