Back on April 18th, @mpopov ping me about a weird paulscore number on dashboard:
Relative paulscore shouldn't be greater than 1. So I checked paulscore_approximations_fulltext_langproj_breakdown.tsv and see:
When F=0.1, the maximum possible score should be 1/(1-0.1) = 1.11111.
Then I forgot about this problem, but it somehow come back to my head today... So I check EL table and see one of those two sessions had two clicks on the first result, the other session had one click on the first result, and hence the paulscores of these two sessions are 2 and 1. This means we should remove duplicated clicks on the same position for each query.
SELECT * FROM TestSearchSatisfaction2_16270835_15423246 WHERE LEFT(timestamp, 8) = "20170326" AND wiki = 'dewikibooks' AND event_source = 'fulltext' AND event_action IN ('searchResultPage', 'click')