Testing and Quality Control Built Into Every Flow
06 eval suites
Features

06.1Contract Suites
Pass or fail assertions check shape and structure with no model call. A case that passed before and fails now marks the run as a regression.
- Define what correct output looks like.
- Check output shape without model calls.
- Catch regressions automatically.
- See which case broke.

06.2Scorer Flows With a Noise Floor
The project’s own scorer flows grade quality. A metric counts as worse only when it moves past statistical noise, and cost changes are reported beside it.
- Grade quality with your own scorers.
- Ignore changes within statistical noise.
- Catch real quality drops.
- See cost changes beside quality.

06.3Rehearse Any Change
Every design write has a dry run. A project plan lists each refusal, cascade and record that would stop fitting, before anything is written.
- Dry-run any change first.
- See refusals before saving.
- See cascades before they happen.
- Find records a change would break.

06.4Dependents Protect Themselves
A flow behind a live endpoint cannot change its inputs or outputs without an explicit opt-in. A table holding records cannot be dropped, and a referenced flow cannot be deleted.
- Keep live endpoints from breaking.
- Change signatures only on purpose.
- Never drop tables holding records.
- Never delete a flow still used.

06.5Checkpoints and Regression Alerts
Snapshot a flow before a risky edit and restore it after. A suite can re-run when its flow changes and raise an event when it regresses.
- Save a flow before risky edits.
- Restore a flow in one step.
- Undo a restore if needed.
- Re-test flows when they change.
- Get alerted when quality regresses.