With this option enabled, training SpamSieve with a good message will add the message sender’s name and address to SpamSieve’s allowlist. This will make SpamSieve classify all future messages from that sender as good. Training SpamSieve with a spam message will disable the name and address if they appear in the allowlist (unless you’ve changed DisableWhitelistRules in the esoteric preferences).
Example
You receive an Amazon order receipt and train SpamSieve with it as a good message. This puts auto-confirm@amazon.com on the allowlist so that future order receipts are always marked as good. A spammer might decide that auto-confirm@amazon.com would make a good fake return address. If you receive such a spam, SpamSieve would mark it as good because the sender was on the allowlist. If you then tell SpamSieve that the message is spam, it will disable the allowlist rule so that it can catch future spam messages with that return address.