I think where the crowd planners are completely off is they assume people are choosing as logically and deliberately as they would. For example... "there won't be a parade here this day and fantasmic runs there and people will be arriving on this day and going here first" type of stuff. While some ppl certainly do think this way, The general tourist just picks and goes. It's easy to assume that because we're diligent about such choices that everyone is... but most ppl are not that way. So, these sites are trying to use an algorithm to predict something that's basically random and pretty distributed anyways. It's like -- studying the lottery picks to predict future picks.
For example... I just got back from a week. Sunday at the MK was not bad at all. Everyone says Sunday is Mother's Day and late EMH so it'll be busy, but it was totally fine.
Throughout the week, the daily weather affected the experience more than anything. Sites recommend against EMH days, but our EMH day we got tons done, and not just in the morning. We pulled Fast Passes all day and by midday we rode the Tea Cups 3 times in a row w no wait.
Consider the 2nd week of May was much rainier than the 3rd week of may this year. We all know rain leads to short lines... so crowd calendars might look back on this next year and say the 2nd week of May is in general better cuz of lighter crowds... when really the crowds were caused by completely random things. Next year it might rain more in the 3rd week.
You can still draw general ideas from them... like on a weekend it'll be busier than on a weekday, and on holiday weekends it'll be really busy... but on a Mon-Fri to say one park is going to be optimal Tuesday vs Wednesday while others are bad is anyone's guess.
And you can also consider this... if everyone did pick days logically, they would all follow the crowd calendars. And if they did that, then the recommended days would be the worst.

So clearly the crowd calculators are predicated on most guests not picking logically to begin with.