Finding bots in a pixel-art community

A first audit of more than 100 million records to find automated accounts.

  • StatusDone once, as a first test.
  • CategoryData and analysis
  • What it does for youMonitors, Reduces errors
Sample report: one sentence says what the check does, three tiles give the totals, and the first pattern is shown with a small chart. Every number is invented.
A sample report built from invented data. No real account appears anywhere.

Sample report

Open a sample report

This sample shows the method. Every number, name and account in it is invented.

Sample data only. No real account appears anywhere.

The problem

On the community site where I am an admin, cheating was found by hand. We had never run an audit, so this was a test.

What I did

I directed an analysis of the full activity history. It looked for patterns that people do not produce:

  • perfectly regular timing
  • accounts that take turns
  • groups that sign up together
  • activity that never rests

Each flagged account got a confidence level and the evidence behind it, so a person could review it.

The result

Some accounts were banned after review. We have done it once so far, and I would do it again.

What this could save your team

The same method finds odd patterns in orders, sign-ups, refunds or reviews.

Sample report: three patterns, each with one sentence and a chart that compares a person with an automated account. Every number is invented.
Three things a person does not do, each shown as a small chart
Sample report, full view: table of flagged sample accounts filtered to high confidence, with the evidence for each one. Every number is invented.
The full view: flagged accounts with evidence and a confidence level

Could a tool like this help you?

Tell me about the task. I will reply with what a tool could do for it.

Tell me your repetitive task