Both native and web applications have a need to experiment, this includes exposing non-standard stats for the sake of evaluating experimental features, or experiments designed to evaluate the usefulness of a new proposed metric in and of itself.

The native part is taken care of at https://crbug.com/webrtc/9410. In Chrome, "is_experimental()" stats must not be exposed by default (this is what happened to legacy getStats() and the standardization process was circumvented). Instead, these should be exposed through Origin Trials.

For reference, I wrote this doc: https://docs.google.com/document/d/17FyH5Oeslk2fF66AOqwUALnDX1bCPrxI8zU3bJXHA08/edit?usp=sharing
But we should re-think how we want experiments to be controlled in Chrome.

To be decided here when we need to run an actual experiment.