It would appear in this case MC isn't actually rounding at all but simply truncating. I would guess its a case of a float being converted to an int and the fraction just being lopped off.
That would certainly
seem to be the case, but what you don't realize is that MC's Group Rating feature is
predictive, not averaged.
You see, MC knows that generally album ratings for any particular artist tend to decline over time, until they finally reach the 0 point. Artists (particularly bands, but it applies to all artists) start out pretty good, since they are in fierce competition to get "noticed" at that time. Occasionally they do initially improve over time, but often these are marginal improvements and they turn out to be short lived. Time is a harsh mistress, and eventually (like with entropy) the sheer power of the vague force known only as Musical Crappiness eventually wins out.
I've taken the liberty of plotting out marko's initial example, adding some additional "predictive" data points to the right end of the chart, showing where this particular band is heading. The X axis represents the length of the artist's career in years, and the Y axis represents the album ratings. Note, when the band first starts off (year 0, if you will), they actually have a negative musical crappiness factor. This, unfortunately, cannot last.... Note the unstoppable direction of the Musical Crappiness Force.
So, while it may seem that your artist should be rated with a 3.75 now... MC knows full well that this rating is going to suffer badly over the next few years, and is acting as an "early warning system" for you. Be thankful! If you monitor the situation closely, you can predict the exact moment when your particular band will have "jumped the shark" and the album won't be worth listening to at all, even once, and avoid making that dreadful mistake. With time, and improved models, MC will be able to fully classify new albums as "buy" or "are you nuts?" with near 100% accuracy.
