Things we want to know include:
- What proportion of logIds come from a browser that says it supports sendBeacon (browserSupportsSendBeacon)? (This should be simple to calculate, and is a top-line figure, but not the most useful figure overall since it is skewed by browser usage and doesn't tell us anything about reliability).
- Of the logIds that says they support sendBeacon:
2a. Have the logEvent, but not logPersistentEvent.
2b. Have logPersistentEvent, but not logEvent
The 2 figures should also have per-browser calculations for all the major browser versions (at least those that are supposed to support sendBeacon, see https://developer.mozilla.org/en-US/docs/Web/API/navigator.sendBeacon )
If we use data from before 30 days after the experiment, the ratio of logged in to anonymous users will be skewed (anonymous users will be slower to start getting the experiment due to HTML caching, but will all have it 30 days after today), but we can still get valid data for individual browsers (e.g. x% of Chrome 33 sendBeacon events came through).
Schema is at https://meta.wikimedia.org/wiki/Schema:SendBeaconReliability .
Results at T101867: Many duplicate events collected from client side javascript