A number like 45% is a claim about a long run of matches, not a forecast of this one.
When we put 45% next to the home win, we are saying that if this same match could be played a thousand times the home side would win about 450 of them. It loses the other 550, and the single result tells you almost nothing about whether the number was right. Hundreds of matches tell you a great deal, which is why the only place we make claims about ourselves is the record page.
A price of 2.10 implies 47.6%. Take a match priced 2.10, 3.40 and 3.60 and the three implied figures are 47.6%, 29.4% and 27.8% — 104.8% in total. The extra 4.8 points are the bookmaker's margin. Scale all three down until they sum to 100% and the match reads 45.4%, 28.1% and 26.5%, which is what we mean by the market.
There is one question worth asking a model that publishes percentages: when it says 45%, does it happen 45% of the time? Take every number in a band and see how often those outcomes occurred. A model can be right more often than the market and still be badly calibrated, which is why we publish the calibration bands.
The market is the aggregate of everyone who has already looked at the match with money at stake, and it updates the moment anything changes. So when our number is a long way from the market's, the honest first assumption is that we are wrong — usually because of team news the model cannot read. Purple marks a disagreement of four points or more, not a verdict. A probability is not a tip. 18+, and please gamble responsibly.