Q: How do I decide which sub-sources to blacklist without overreacting?

Q: How do I decide which sub-sources to blacklist without overreacting?

A: Blacklisting is surgery, not a sledgehammer. Run this process per sub-source:

— Gather 7 days of data minimum. One bad day isn't a verdict.
— Score each sub-ID on three numbers: rejection rate, fraud score, and conversion-to-payout rate.
— Flag, don't ban, the outliers first. Move them to a watchlist.
— Throttle the watchlist to a small cap and re-measure for 3 more days.
— Confirm the cause: was it bots, geo, or just a weak audience? Weak isn't fraud.
— Blacklist only sources that stay bad after throttling AND show fraud signals, not just low conversions.
— Document the reason so you can defend the decision later.

Short version: 7 days of data, score three metrics, watchlist, throttle, then ban only proven-bad sources.

Still stuck? Drop your case in the comments.