Running and past tests are now separate on the Experiments page

The Experiments page has the same layout on its three tabs.

• Running tests has a card for each test running now. The section is hidden when nothing runs.
• Past tests lists the finished tests. Filter them by Result and by Started between. There is no status filter, since running tests have their own section.
• New test, at the top right, starts a test. On the Widget and Agent tabs it opens a dialog where you pick the widget or agent to test and start the test, without leaving the Experiments page. A widget that is not ready to test says why and links to its editor. An agent that is not ready lists the start checks in the dialog, met or not, and has no link. On the Holdout tab it opens the dialog that starts the holdout test. The button needs the permission to edit widgets, agents or sites, depending on the tab.
• A tab with no test yet shows the same empty state: No widget test yet, No agent test yet or No holdout test yet.

An agent that has a test now has its own page under Experiments to read it: the versions, the verdict, the live reading, the secondary drivers and the limits, then the stop, promote or keep buttons. Running and past agent tests link to it. The Experiment tab of the agent editor stays and works as before, with its readiness card and start button.

In English, the third tab is now called Holdout instead of Increment. The French and German labels do not change, and the tab links still work.

On the Holdout tab, the start button no longer sits inside the card. Use New test, which shows while no holdout test runs.