build

Checks the vendor before it writes code

It deploys a sandbox and puts a real request to the twin for each thing the design assumes: what a create returns, whether a field is nullable, what happens on a repeat, which error a refusal produces. For a feature about a failure path, it arms the fault and drives your current code through it first.

It also varies whatever the design uses to identify a record and counts the rows the twin stores, so distinct inputs that collapse into one record show up here.

Writes the feature

It follows your repository’s existing conventions, then runs your full test suite.

Proves the new path against the twin

The new code runs against the same sandbox the checks ran against, so it meets the state and any armed faults they used. On the container tier that run is veris run and its receipt is the record; on the direct tier the record is the twin’s trace, read from the id noted before the run. The work is not done until that record shows the twin was called.

Opens a PR

Three sections: what was verified and how, what was assumed rather than verified, and limitations and risks. It names the sandbox id, and other entry points that reach the same behaviour are either driven too or listed as a risk.