Research talk:Reading time/Work log/2018-10-23

Add discussion
Active discussions

Tuesday, October 23, 2018Edit

Notes from meeting with JonEdit

  • If visible length is not supported in the browser then visible length == total length. We can identify browsers that don't support visible length by looking for browsers where that's always the case.
  • Quick Back button behavior could explain page unload view without page unloaded.
  • You should have one page token per user session.
  • Group by page token and by user token because of collisions.
  • DONE: share "page unloaded before page loaded" cases with Jon and nearby events.
  • If page token collisions are not super rare we have a problem.
  • Cases with multiple page unload event might be interrupted edits. Reading depth seems to run on edits? DONE: file a bug about that (mark all action pages).
  • We might have cases on mobile with no page unloaded event, and we won't get the view. So we are likely to have missing data on mobile. We can measure this by counting the number of page loaded events without page unloaded events. Done: run this and update Jon and Tilman ( work log).
  • Make sure that removing IOS < 11.3 removes almost all the Mobile Safari views.

Tilman Bayer3:09 PM

Jon Robson3:11 PM

Thanks for these notes! I guess something got mixed up regarding "You should have one page token per user session" - different pageviews within one session can and should have different page tokens (please file a bug if you encounter a case where this isn't so). And "Group by page token and by user token because of collisions" should not be necessary - page token collisions should not just be super rare but virtually impossible in this dataset, considering that we increased their length to 80 bits recently (and that the ReadingDepth schema had a much lower number of events before that change). Regards, Tbayer (WMF) (talk) 22:15, 24 October 2018 (UTC)
Return to "Reading time/Work log/2018-10-23" page.