It was mainly Mikey, although I was paying a fair amount of attention to it too.
As it happens, ickieria's being more than a little harsh to call the site "a load of "b*ll*cks", and TTT appears to be demonstrating his usual nuanced understanding of political science. In the first place, electoralcalculus is not a polling organisation. Instead it took the outputs of opinion polls published by other organisations and fed those numbers into a model to produce predictions based on individual constituency results. Problem is that the polls turned out to be off by far enough to make the model sophistication largely irrelevant: "Garbage in, garbage out", which is a common enough phrase in computer programming. The model used by the site was tested after the election using the actual results and came up with a new 'prediction' that was correct in predicting the seat results for 624/650 seats, with 21 of the wrongly-predicted seats having majorities of less than 5%. Given TTT's favourite comment about our current system, that percentages do not equal seats, getting such a large proportion of the constituency results correct based on national percentages suggests quite a decent model, actually. The problem electoralcalculus, and everyone else, faced was that the polling was so badly wrong, but that wasn't really in their hands.