# Write the brief and the pass rule

Lesson 1 of [Run a GPU cloud trial before you commit](https://gpuclouds.fyi/academy/run-a-gpu-cloud-trial.md), 12 minutes.

Start with the result you need. A trial without a pass rule only tells you that something ran.

## Record the workload

- For training: the model, the precision, the batch size, the data path and the checkpoint size.
- For inference: the request sizes, the concurrency and the latency you must meet.
- For both: the region you need and how sensitive the data is.

Do not use a vendor's peak throughput as your application benchmark. Measure your own workload.

## Write the pass condition first

Write the pass condition before you run anything. Include these items:

- Correctness. Write the expected output before you start, so that a fast but wrong run cannot pass.
- Runtime.
- Maximum spend.
- Recovery from an interruption.

## Start small

Choose a small workload with a known result. Use public or harmless data for the trial. Expand only after the run is repeatable.

## Check capacity early

Check that the device is available in the actual region you need. A public device catalogue does not reserve capacity. Ask for the quota and the allocation terms before you plan a launch.

## Keep a record of rejects

Keep a note of each rejected candidate and the condition it failed. This helps when prices or capacity change.

## Sources

- GPUClouds guide: Choose a GPU cloud for your workload
- GPUClouds guide: Prepare your first GPU cloud run

Also on the [web page](https://gpuclouds.fyi/academy/run-a-gpu-cloud-trial/write-the-brief-and-pass-rule).