In T354597: Record IP reputation data for account creations and edits we created a schema and server-side instrumentation to record IP reputation data obtained via Spur.us / iPoid-Service, if it exists, for a given IP when edits and account creations occur.

In this task, we want to answer the following questions:

1. Is there a correlation between specific attributes of IP reputation data and reverted edits?
   1. When productive edits occur from IPs with poor reputation, which IP signals are present or absent?
2. Is there a correlation between specific attributes of IP reputation data and blocked accounts?
3. Is there a correlation between negative IP reputation data and anonymous editing compared to logged-in editing?
4. What percentage of reverted edits are associated with IP reputation data obtained from Spur / iPoid-Service ?
5. What percentage of blocked (one-off, or permanently) accounts are associated with IP reputation data from Spur / iPoid-Service?
6. In general, can we explain the mapping between IP signals and on-wiki editing by a reduced set of attributes?
7. Which infrastructure, client, risk, platform, protocol, client behavior, service category, targeting type enums and tags are most problematic in terms of being associated to reverted/deleted edits on wiki and permanently blocked accounts? Which are least problematic? https://docs.spur.us/data-types?id=data-types

Notes

Reverted Edits

• We are interested in content edits on English Wikipedia.
• Use 48 hours as the cutoff for revert status.

Blocked Account

• We are interested in accounts created in English Wikipedia, excluding auto account creations.
• Blocks include both global blocks and local blocks in English Wikipedia.
• Blocks will not break down by duration (permanent vs. temporary) since most blocked registered accounts (97%) in the sample have been issued a permanent block.
• We are primarily interested in < 90 days old accounts, assuming that accounts with persistent bad faith activity older than that will have already been blocked. Plus our events data is only available for 90 days.