Page MenuHomePhabricator

Consider recalculating revert rate
Open, MediumPublic

Description

The revert rate we use in our Key Product Metrics is problematic. Currently, it's:
Reverted edits (including reverted bot edits) / Non-bot edits

https://www.mediawiki.org/wiki/Wikimedia_Product/Data_dictionary#Core_metrics

The definition was made under the assumption that bot edits are generally not reverted, but we've found that assumption to be incorrect.

In addition, we may want to consider only calculating reverts done in a set window after original edit. @nettrom_WMF has used 48-hour revert rate in Growth.

Event Timeline

kzimmerman updated the task description. (Show Details)Nov 2 2020, 6:32 PM
kzimmerman added a subscriber: nettrom_WMF.
cchen added a subscriber: cchen.Nov 5 2020, 6:45 PM
kzimmerman triaged this task as Medium priority.Nov 5 2020, 6:51 PM
kzimmerman moved this task from Triage to Upcoming Quarter on the Product-Analytics board.

When we tackle this, we should loop in Analytics Engineering and consider whether we need new calculations in mediawikihistory.

We should also consider implications from T266374 -- differences between checksum-based and revert-tag based reverts in mediawiki_history

fdans edited projects, added Analytics-Radar; removed Analytics.Nov 9 2020, 4:30 PM