Assessments and runs
Understand assessment planning, run states, test checks, and verified outcomes.
An assessment applies an approved journey and its test cases to one or more active agent environments. A run is one execution record within that assessment.
Before a run starts
Invokeable confirms that:
- The journey has an approved, runnable test plan.
- Required product actions are ready.
- At least one selected agent environment is active and compatible.
- The connected environment is authorized for the requested action type.
- Time, action, spend, and concurrency limits are available.
- The expected result can be checked independently.
Run lifecycle
| Phase | What is happening |
|---|---|
| Created or queued | The plan exists and is waiting to start |
| Dispatching | Invokeable is preparing the execution environment |
| Running | The agent is working through the customer task |
| Waiting | The run is paused for a defined delay or retry condition |
| Verifying | Invokeable is checking product state and evidence |
| Cleaning up | Temporary state is being restored or removed |
| Completed | The run reached a terminal state and has an outcome |
| Canceled or stopped | Work ended before ordinary completion |
| Failed or indeterminate | Execution or evidence could not support a trustworthy result |
Cancel applies to work that has not started. Stop requests a safe end to active work. Either action may require cleanup before the run becomes terminal.
Checks and outcomes
Each test case contains checks for behavior such as action selection, valid inputs, sequence, final state, safety, recovery, communication, and efficiency.
A check can be:
- Passed when the expected behavior has supporting evidence
- Failed when evidence shows the expected behavior was not met
- Inconclusive when evidence is missing, conflicting, or unavailable
The customer-facing outcome summarizes those checks. See Statuses and outcomes for exact meanings.
Repeated trials
Some failures appear only under timing, model, or environment variation. Repeated trials help measure consistency and distinguish a reproducible problem from a one-off limitation. Invokeable keeps every trial's evidence and records material exclusions.
Scheduled assessments
Assessment plans can run immediately or on a schedule. Scheduled work still uses the approved journey, action versions, agent environments, and safety limits captured by the plan. A schedule never bypasses readiness or authorization checks.
Continue with Run an assessment or Evidence and findings.