A workflow defines a sequence of actions. An execution is one run of that workflow with particular input values.

The distinction matters when you check a result. The same workflow can receive different inputs and take different paths on each run.

## The definition is not the result

Consider a client request app. These names and rules are an example, not a built-in workflow:

- A form submission starts the workflow.
- A Branch step checks whether the request has a title.
- A Data step saves a valid request.
- An Integration step sends a receipt through Resend.

The workflow describes those actions. One submission for *Replace the lobby sign* is one execution. A later submission is a different execution.

A workflow definition can look correct while a particular execution fails. For example, a request can reach the email step but fail to produce a receipt.

The saved request is a [record](/concepts/entities-and-records), not the workflow itself. The record and the email are separate results to check.

## Inputs determine the path

Each step does one job and passes data to the next step. A Branch step chooses a path based on a condition.

In the example, a request with a title should reach the save step. A request without a title should follow the other path.

Those are two cases for the same definition. A successful test with a title does not show what happens without one.

Use the [Workflow workspace](/features/workflow-workspace) to inspect the steps that make up the definition.

## Simulation shows one run on example data

**Simulate** runs the workflow on example data. Open a step to see its inputs, output, and duration.

For the request example, inspect whether the title reaches the save step and what the save step passes to the email step. Use a separate case without a title to examine the other branch.

Simulation gives evidence about the tested inputs. It does not prove that every possible request works, or that a receipt reaches a real inbox.

## Live executions run on a server

A live workflow runs on the app's server, even when your browser is closed. Closing Studio does not stop that server-side work.

The [Orbit workflows overview](https://orbitapps.io/platform/workflows) states that failed workflows return to a durable queue for retries. Orbit waits longer between tries, up to twenty retries. Orbit logs a run that cannot finish with its error.

A retry is another attempt to complete work. It is not proof that the intended result exists. The overview does not promise that earlier external actions are reversed after a later failure.

For the request example, do not assume that an email error also means the request was not saved. Inspect the saved request and the recipient's inbox separately before you submit the request again.

## Check the outcome people need

A step output helps explain a run. The app's final result answers a different question: did the person get what they needed?

For this example, the result checks are:

- The request appears in the app with the submitted title.
- The request remains after you reopen the app.
- The receipt arrives in the recipient's inbox.
- A request without a title follows the intended alternative path.

[Test the app before launch](/guides/review-an-app-before-launch) with the accounts and services people will use. Keep the input case, step results, and final outcome distinct when you investigate a failure.