I believe I have asked this before, but complaints from my users have prompted me to revisit the topic. From the perspective of myself and the users, the spam learning filter seems to do nothing. I have had several users report that they continue to receive certain repeat spam messages every day despite always clicking the "Spam" button. I myself have experienced this as well. To give an example:

X-Spam-Flag: NO
X-Spam-Score: -4.897
X-Spam-Status: No, score=-4.897 tagged_above=-10 required=6.6
tests=[AM:BOOST=-3, BAYES_00=-1.9, DKIM_ADSP_CUSTOM_MED=0.001,
T_RP_MATCHES_RCVD=-0.01, T_TO_NO_BRKTS_FREEMAIL=0.01] autolearn=ham
This is from a message I receive every day from a Yahoo Groups spam, which I never signed up for and have been unsuccessfully able to remove myself from. It's always from the same address, and the message is always formatted the same way, so it seems it would be an easy task for the learning mechanism to recognize it. I have flagged it as spam every day for at least 2 months now, but as you can see from the header it still shows a Bayesian probability of 0. At this point I've completely lost faith in the "Spam" button. Or am I misunderstanding what it's supposed to do?