Currently, we're running the calibration script on our virtual workstation. We're seeing stability in results between runs, but we've noticed that the results can differ depending on the time of day we run the script.

We should run the script on a machine which minimises noise and leads to reproducible coefficients. The performance team has given a few ideas:

• Change host type to amazon2023.3-arm64-large
• Try different AWS zones for the host
• Get a dedicated AWS instance
• Try to run script in DSI and enable settings such as dedicated tenancy
• Do the infrastructure provisioning DSI step to provision new AWS instance, then tear down + re-provision + repeat
• Use bare metal machine (if nothing else works)