Build beyond the first check
Once you can validate a model and explain a failed population, choose the next guide by the job your application needs to do. Each workflow produces a specific artifact or result that you can inspect.
| Your next question | Guide |
|---|---|
| How do I translate requirements into checkable rules? | Modeling |
| How do I derive eligibility, use a fixed catalog, or audit a change? | Data and transitions |
| How do I generate application structure and detect drift? | Generated projects |
| How do I exercise rules with synthetic cases? | Synthetic data |
| How do I change existing rules or recover a model from code? | Evolution and recovery |
| How do I check facts extracted from a report? | Documents |
| How do I run a bounded agent experiment? | Experiments |
| How do I explain a derived tuple or use a native export? | Exports and witnesses |
| How do I ask an agent to use these workflows? | Agent skills |
For a continuous learning path, build BookRec, then try the expense application. The capability matrix shows which features compose and which consumers have narrower support. The CLI index and Python API reference help you integrate the parts you select.
The examples use synthetic public inputs and deterministic local checkpoints. Saved proposals, transcripts, receipts, and replay artifacts retain their original scope; replaying them is not evidence that a provider ran now. Live agents, advisory reviewers, and native compilers have separately selected prerequisites. A mechanical pass is distinct from formal assurance: missing or stale evidence yields unavailable assurance. These tutorials do not recertify changed bytes or claim a measured provider/native support profile.