@JGirault noticed a discrepancy between the clickthrough rate on the Portal dashboard and the rate reported in the most recent Portal A/B test analysis.

The only difference is that the data used in the A/B test report underwent an additional cleaning step wherein duplicated events were removed. Therefore the clickthrough rate that is surfaced on the Portal dashboard is calculated using faulty data that has A LOT of duplicate events (any session should have at most 1 landing event and 1 clickthrough event).

An example of faulty data:

Here are the top 10 sessions from 2016-04-25 by number of events per session:

USE log;
SELECT session, type, COUNT(1) AS n_events FROM (
  SELECT event_session_id AS session, event_event_type AS type
  FROM WikipediaPortal_14377354
  WHERE LEFT(timestamp, 8) = '20160425' AND ((event_cohort IS NULL) OR (event_cohort IN ('null','baseline')))
) AS events
GROUP BY session, type
ORDER BY COUNT(1) DESC
LIMIT 10;

That is not good and should be corrected in the nearest future.