= Potential problems =
* Sessions with many init events?
= Data =
== Count of events, May–July 2018 ==
| | visual: desktop | visual: other | visual: phone | wikitext: desktop | wikitext: phone | 2017 wikitext: desktop | 2017 wikitext: other |
|--------------|--------------|--------------|----------|----------|---------------|---------------|-------|
| init | 28735 | 860555 | 32601 | 0 | 3814652 | 7139 | 30409
| loaded | 818006 | 0 | 0 | 4608821 | 0 | 36911 | 0
| ready | 817956 | 0 | 49765 | 4633070 | 3682677 | 11989 | 24507
| abort | 699164 | 0 | 3481 | 2701114 | 566641 | 12895 | 0
| saveIntent | 130588 | 0 | 9757 | 0 | 148291 | 33710 | 0
| saveAttempt | 134481 | 0 | 9246 | 0 | 111767 | 24031 | 0
| saveSuccess | 107314 | 0 | 6624 | 0 | 92504 | 23451 | 0
| saveFailure | 28628 | 0 | 0 | 0 | 18131 | 778 | 0
== Count of phone events by month ==
| month | wikitext events | visual events
|---------|-----------------|---------------
| 2017-12 | 2037604 | 28271
| 2018-01 | 4341238 | 59143
| 2018-02 | 3897444 | 53763
| 2018-03 | 4220510 | 59607
| 2018-04 | 4084998 | 57519
| 2018-05 | 4263057 | 49500
| 2018-06 | 4170920 | 61958
| 2018-07 | 686 | 16
| 2018-08 | 66 | 0
= Actual problems =
* Large number of VE `init` events with platform `other`, which seem to be misclassified `desktop` events
* Large number of 2017 WTE `init` and `ready` events with platform `other`, which seem to be misclassified `desktop` events
* No `loaded` or `saveFailure` events recorded for mobile VE
* No `init` or `save` events recorded for `desktop` WTE
* No `loaded` events recorded for the `phone` WTE
* Possibly disproportionate number of `init` and `ready` events from the `phone` WTE
** This could just be a high abort rate, but then why aren't we seeing more `abort` events? `abort` + `saveAttempt` << `ready`
* Phone events (both visual and wikitext) almost entirely stopped in late June or early July 2018
** @Jdlrobson may be taking care of this in T202786
* Visual editor uses 16 character (64 bit) session tokens, while the wikitext editor uses 32 character (128 bit) ones.
** This doesn't affect the data quality, but it just feels weird to me (Neil). I think I remember that client-side events don't use the pre-generated pageview token, so perhaps that would be a quick fix.