Would this release ship?

Explore how a test pipeline uses AI advice while a deterministic release policy decides what ships.

Replay a release

Recorded test evidence

Loading the scenario evidence…

Can an AI draft catch a missed bug?

Recorded coverage challenge

The existing API suite caught 1 of 2 injected faults. Adding the qualified draft raised that to 2 of 2.

Checkout fixture	Existing API suite	Suite + AI draft
Healthy checkout	Passed	Passed
Ordinary retry creates a duplicate charge	Caught	Caught
Retry after a lost reply creates a duplicate charge	Missed	Caught

Retry returned order-2 instead of the committed receipt order-1

Inspect the drafted test

llama3.2:latest selected a withheld first reply and two concurrent retries. The harness checked that each retry returned the same paid receipt. The prompt contained the API contract and allowed operations.

How reliable was the AI advice?

Recorded benchmark

Loading benchmark results…