The original title is about making chunked upload finalization idempotent before network retries. Let me rewrite this for a mobile feed.

Short summary

Chunked upload APIs that lack idempotency can create duplicate artifacts or corrupt data when clients retry lost finalize responses. The post prescribes server-side chunk hash verification, compare-and-set state transitions from uploading to complete, and returning the same artifact ID on retry. It covers duplicate chunk handling, missing-index errors, expired session cleanup, and end-to-end test coverage across UI, API, storage, and assembly layers.

• •Chunk PUTs must be idempotent for identical bytes and reject conflicting hashes with 409
• •Finalize uses a compare-and-set state machine returning the same artifact ID on retry
• •Sweeper cleanup must avoid removing sessions that are mid-finalization

Generated with AI, which can make mistakes.