KnitKnot

# Run and schedule benchmarks

How to trigger a benchmark manually, put it on a recurring schedule, and read benchmarks as the periods that drive your score trend.

A **run** sends your active prompt library to the engines selected for that run. KnitKnot supports ChatGPT, Claude, Perplexity, and Gemini; the choices available to your workspace depend on its plan. Results from scored runs are grouped into benchmark periods so you can inspect a reading over time.

Open [Benchmarks → Results](https://app.knitknot.ai/benchmarks) to run or schedule the benchmark. Use **Benchmarks → Runs** to inspect batches and their answers.

## Trigger a run

Use **Run all**, select the available engines, then select **Run**. The work continues in the background. **Runs** shows progress and the answers behind a batch. When scoring finishes, **Results** shows the benchmark period. If some answers failed, check the run status before treating the period as a full reading.

Trigger a manual run after you've shipped a fix and want to check whether AI has picked it up, or any time you want a fresh reading.

## Put it on a schedule

Use **Schedule** on the Benchmarks page to set a recurring run. Choose **weekly**, **biweekly**, **monthly**, or a **custom** cadence, or turn the schedule off. Check the selected engines in the schedule before saving it.

The scheduled run uses the active prompt library when it starts and the engines saved with that schedule. To compare it with a manual run, check that both measured the same prompts and engines.

## Runs, periods, and your trend

Runs group into periods. Open a period to inspect **Changes**, **Head-to-head**, **Runs**, and **Reports**. A period with missing answers can still show a score based on what was measured, so check its coverage before comparing it with another period.

Run again when you need a new observation, such as after publishing a change. A later answer can differ because the engine or its sources changed, even if you did not change a page. Keep the prompt and engine scope steady when the purpose is comparison.

Next: once a run completes, [read your report](/docs/read-your-report/) and work through the [issues and playbooks](/docs/issues-and-playbooks/) it surfaces.

Docs navigation
Docs › Guides

Run and schedule benchmarks

How to trigger a benchmark manually, put it on a recurring schedule, and read benchmarks as the periods that drive your score trend.

Updated

A run sends your active prompt library to the engines selected for that run. KnitKnot supports ChatGPT, Claude, Perplexity, and Gemini; the choices available to your workspace depend on its plan. Results from scored runs are grouped into benchmark periods so you can inspect a reading over time.

Open Benchmarks → Results to run or schedule the benchmark. Use Benchmarks → Runs to inspect batches and their answers.

Trigger a run

Use Run all, select the available engines, then select Run. The work continues in the background. Runs shows progress and the answers behind a batch. When scoring finishes, Results shows the benchmark period. If some answers failed, check the run status before treating the period as a full reading.

Trigger a manual run after you’ve shipped a fix and want to check whether AI has picked it up, or any time you want a fresh reading.

Put it on a schedule

Use Schedule on the Benchmarks page to set a recurring run. Choose weekly, biweekly, monthly, or a custom cadence, or turn the schedule off. Check the selected engines in the schedule before saving it.

The scheduled run uses the active prompt library when it starts and the engines saved with that schedule. To compare it with a manual run, check that both measured the same prompts and engines.

Runs, periods, and your trend

Runs group into periods. Open a period to inspect Changes, Head-to-head, Runs, and Reports. A period with missing answers can still show a score based on what was measured, so check its coverage before comparing it with another period.

Run again when you need a new observation, such as after publishing a change. A later answer can differ because the engine or its sources changed, even if you did not change a page. Keep the prompt and engine scope steady when the purpose is comparison.

Next: once a run completes, read your report and work through the issues and playbooks it surfaces.