> ## Documentation Index
> Fetch the complete documentation index at: https://docs.alethiaai.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Scheduled & Recurring Tests

> Run tests automatically — once at a future time, or on a recurring cadence.

You don't have to be at the keyboard for tests to run. Alethia supports two kinds of automation:

<CardGroup cols={2}>
  <Card title="One-time scheduled test" icon="calendar-plus">
    Pick a single date and time in the future. Alethia runs the test then.
  </Card>

  <Card title="Recurring schedule" icon="repeat">
    Pick a cadence (daily, weekly, monthly). Alethia keeps running the same test on that schedule until you stop it.
  </Card>
</CardGroup>

Both are managed from the **Schedules** tab inside any project.

## Why schedule tests?

* **Catch regressions automatically.** Providers update models without notice. A weekly run on the same prompts catches behavior changes the moment they happen.
* **Build a compliance trail.** A monthly recurring run on your high-risk prompts is exactly the kind of "continuous risk monitoring" the EU AI Act expects.
* **Save engineering time.** Set it once, forget it, review the results when you need them.

## Creating a one-time scheduled test

<Steps>
  <Step title="Go to the Schedules tab">
    Inside your project.
  </Step>

  <Step title="Click 'Schedule a Test'">
    Choose **One-time**.
  </Step>

  <Step title="Pick what to run">
    A single prompt or an entire category.
  </Step>

  <Step title="Pick when">
    Date, time, and timezone.
  </Step>

  <Step title="Save">
    The schedule appears in the **Pending** list. You can edit or cancel it any time before it runs.
  </Step>
</Steps>

When the time hits, Alethia kicks off the test run. Results appear in the normal **Results** tab, tagged with the source schedule.

## Creating a recurring schedule

<Steps>
  <Step title="Click 'Schedule a Test'">
    Choose **Recurring**.
  </Step>

  <Step title="Pick what to run">
    A single prompt or an entire category. Recurring schedules are great for category tests — they give you week-over-week coverage data.
  </Step>

  <Step title="Pick the cadence">
    * **Daily** — every N days at the time you choose
    * **Weekly** — pick the days of the week
    * **Monthly** — pick a day of the month (1–31; the schedule auto-skips months that don't have that day)
  </Step>

  <Step title="Pick the time of day">
    24-hour format. Pick a timezone (defaults to UTC).
  </Step>

  <Step title="Set start and (optional) end">
    * **Starts at** — when the first run happens (defaults to now)
    * **Ends at** — when to stop, *or* a maximum number of occurrences
  </Step>

  <Step title="Save">
    The schedule shows up in the **Active** list with the next run time.
  </Step>
</Steps>

<Tip>
  **Pick off-peak times.** Provider APIs are faster and less rate-limited at night. A daily run at 03:00 UTC tends to complete much quicker than one at 14:00 UTC.
</Tip>

## Managing schedules

In the **Schedules** tab you can:

* **Pause / resume** a recurring schedule (toggle the switch)
* **Edit** the cadence, time, or end conditions
* **Cancel** a one-time schedule before it runs
* **View history** — every run a recurring schedule has produced, linking back to the test results

## What if I edit the prompt while a recurring schedule runs?

Each individual scheduled run uses **the prompt as it exists at the moment of execution**, not as it was when the schedule was created. So if you update a prompt's content tomorrow, future scheduled runs will use the new content.

Past test results are unaffected — they keep referencing the prompt as it was when they ran.

## What if a scheduled run fails?

Failed runs appear in the **Results** tab with a red status. The recurring schedule keeps going — a single failure doesn't pause the schedule. Common failure reasons:

* The provider's API was down at the moment of execution
* The API key has been rotated and isn't updated in the LLM Configuration
* A judge model has been deprecated

Alethia retries each call up to 3 times with exponential backoff before marking the test failed.

## How does this fit with EU AI Act compliance?

Article 9 of the EU AI Act requires a "**continuous, iterative risk management process**". Recurring scheduled tests are how you do this in practice:

* A weekly run on safety-critical categories produces ongoing evidence of monitoring
* The test history (visible in Schedules → History) is the evidence trail
* Each run can feed into a periodic [risk assessment](/risk-management/assessments) automatically

For a high-risk AI system, a typical setup is:

* **Daily** smoke test on a small set of critical prompts
* **Weekly** full run of all categories tied to the system's risk register
* **Monthly** post-assessment to summarize trends and update the risk register

## Limits

| Limit                                           | Value              |
| ----------------------------------------------- | ------------------ |
| Maximum recurring schedules per project         | 50                 |
| Maximum scheduled runs per day per organization | Set by your plan   |
| Schedule timezone precision                     | 1 minute           |
| Lookahead window for one-time schedules         | Up to 1 year ahead |

If you hit a limit, contact support or upgrade your plan from **Organization → Billing**.