followup on pranks

A few months ago, I wrote that we would be using a 'prank' queue to identify patrons coming from IP addresses that had been associated with the 'prank' resolution code in the past hour.

You can read up on that link to find out how it works and why, but the main idea is to save librarians from the trouble of dealing with repeat pranksters, or at least to warn them.

I checked today and found out, since we added the queue system on March 12, librarians have used the Prank resolution code 310 times.

+--------------+-------+
| resolution   | count |
+--------------+-------+
|              |    16 |
| Completed    |  3570 |
| Disconnected |   740 |
| Followup     |   465 |
| Prank        |   310 |
| Test         |    33 |
| Transferred  |    68 |
+--------------+-------+

Of those, 61 came in the 'prank' queue, meaning they were repeat pranksters within an hour. That doesn't seem very significant - 61 of 310 questions (20%) coded 'prank' came within the hour of another from the same IP address.

+----------+--------+
| queue    | pranks |
+----------+--------+
| academic |      3 |
| lnet     |    246 |
| prank    |     61 |
+----------+--------+

Keep in mind we sometimes questions don't export right after they are finished, so some repeats might have snuck in on the regular queues before we got the resolution code from the first session. I am going to call this at least 20%.

I also looked at the resolution codes used for questions in the 'prank' queue.

+--------------+-------+
| resolution   | count |
+--------------+-------+
| Completed    |    98 |
| Disconnected |    42 |
| Followup     |     1 |
| Prank        |    61 |
+--------------+-------+

Almost half of the questions coming into the prank queue were coded 'Completed', indicating that they weren't questions from repeat pranksters after all. The percentage of the time they were coded pranks is significant though - 249 of 5,000 (5%) of questions in the 'lnet' and 'academic' queues were coded prank, compared with 61 of 202 (30%) in the 'prank' queue. There is also a higher rate for the 'Disconnected' code, 20% as opposed to 15%.

There are three ways we can go from here:

• Leave this system in place
• Scrap this system
• Adjust the parameters of how we define who should go in the 'prank' queue. We could change the amount of time we label an IP address, look at other parameters, etc.

My main question for L-net staff is, does it help you to know that a patron might be a prankster?