• How many chat messages have profanity? Hate speech?

• Are we filtering them effectively?

• Have we optimized the detection and filtering algorithms?

• Can we (should we) pay for a service for this, or do it ourselves?

To answer these questions, let's

• "replay" chat messages from two days ago

• feed them through various detection and/or filtering systems

• collect whatever metrics we want (including performance)

Over three days...

In the top 2 charts on the right, we can see metrics around fetching the chat logs from BigQuery. Then they are fed into a Reactive Flux. (See LogStreamService)