Hum, i forgot one thing. I knew i was forgetting something, but it came back 2 minutes _after_ sending the previous mail, as always #@!
=== Antispam ===
The antispam system is far from being used by everybody, and it's a lot of ressource consumption. So the default mode for training has been changed to TUM (train until mature) ; explanation follows.
The previous mode, TEFT (train every fucking time), was meant to do an intensive session training, until the system has learnt enough to use a more relaxing mode. So for the begining we thought it was a good decision, but in fact if people don't use it, it consummes a lot of ressource for learning bullshit.
Once the system has trained a lot, you may consider using TOE mode (train on error). It is good, and requires less user attention. But the drawback of this mode is: it is less reactive to new spam types, as it does not recognize new prints, but wait for an error being corrected by the user.
That's why the TUM mode has been choosen as an intermediate solution: as reactive as TEFT for new prints, but not much demanding for training older stuff, like in TOE mode.
If you already made changes to your preferences, nothing changed, and you may or may not decide to follow this solution. Only people using the default parameters are concerned.