Historically, experiments have been enabled/disabled by changing custom config variables defined in the operations/mediawiki-config repo. The status and history of an experiment has been captured in config variables, Git history, and the Server Admin Log. Deployments are also visible in Grafana, meaning that if a metric has been impacted, then someone would be able to at least tie it back to a deployment.

The Experiment Control Plane removes the friction from enabling/disabling an experiment but at the cost of adding somewhere else to look to determine the root cause of an impacted metric.

TOINVESTIGATE

• Who should be informed when an experiment is enabled/disabled?
• What tooling is available to notify them from the application servers?
  • e.g. Can we create Grafana annotations from the application servers?