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

# Scheduled Evaluations

Scheduled evaluations run **automatically on a recurring schedule**, so you don't have to remember to test an agent by hand. On each scheduled run, VIVI samples the agent's recent **live conversations** and scores them — giving you a continuous, hands-off read on quality as real traffic changes over time.

<Note>
  Scheduled evaluations are **Live-source only** — they grade a sample of real recorded conversations, not uploaded datasets. You can't schedule a dataset run.
</Note>

***

## Creating a schedule

Go to **Evaluations → Scheduled** and click **Schedule Evaluation**.

<Steps>
  <Step title="Select an agent">
    Choose the agent to evaluate on a recurring basis.
  </Step>

  <Step title="Set the sample size">
    Enter the **Number of conversations** to score on each run.
  </Step>

  <Step title="Add an expected outcome (optional)">
    Describe the goal a conversation must achieve to count as successful. If you skip this, Correctness is skipped on each run.
  </Step>

  <Step title="Describe the schedule">
    Type the cadence in plain language, for example **"every day at 9 AM"** or **"every Monday at 2:30 PM"**.
  </Step>

  <Step title="Confirm the timezone">
    Set the timezone so runs align with your business hours. It defaults to yours.
  </Step>

  <Step title="Save">
    The schedule appears in your list with its cadence, timezone, and next run time.
  </Step>
</Steps>

***

## Schedule limits

Open the **Schedule limitations** popover on the panel for the full list of guardrails:

<Warning>
  * Runs **at most once per day** (at least **24 hours** between runs).
  * Runs **at least once every 14 days**.
  * Times are set in **30-minute steps** (e.g. 12:00, 12:30, 1:00).
</Warning>

**Examples that work:** *"every day at 9 AM"*, *"every Monday at 2:30 PM"*, *"every 3 days at 8 AM"*, *"every weekday at 6 PM"*.

VIVI validates your schedule as you type and explains anything it can't accept:

| Message                              | What to change                                        |
| ------------------------------------ | ----------------------------------------------------- |
| "Use a specific recurring time…"     | Make an ambiguous description more specific.          |
| "Couldn't understand this schedule…" | Rewrite a schedule VIVI couldn't parse.               |
| "…at most once every 24 hours"       | The schedule runs too often — space it out.           |
| "…at least once every 14 days"       | The schedule runs too rarely — make it more frequent. |
| "…can't be in the past"              | Choose a future recurring time.                       |

***

## Managing schedules

The Scheduled table shows each schedule's **Agent**, **Cadence**, **Timezone**, **Next Run**, and when it was **Created**, with actions to **Edit** or **Delete**.

* On each due tick, VIVI **creates a new live run** under the schedule — you'll see it appear on the Runs page like any other run.
* **Editing a schedule** only changes future runs; your **historical runs stay intact**.
* If a scheduled run finds **no conversations** to evaluate, it's marked **Skipped** — a benign status, not a failure. There was simply nothing to score.

<Note>
  A skipped scheduled run usually means the agent had no recorded conversations in the last 14 days. Once activity resumes, the next scheduled run will pick them up.
</Note>

***

## Best Practices

* Set a **regular cadence** (for example, weekly) for every agent that's in production.
* Choose a **sample size** large enough to be representative, but stay under the 500-evaluation run limit.
* Pair a recurring schedule with [on-demand summaries](/documentation/features/evaluations/summaries) so you can compare a recent window against the agent's history.
* Re-check the **timezone** when you roll out to a new region — a wrong timezone samples the wrong part of the day.
* Review your schedules from time to time to confirm they're still the right agents and cadence.