Not that it really matters, but Electoral Calculus are not pollsters. They take national polling data that actual polling companies release and then process it in order to generate a prediction for how the raw polling numbers will translate into actual constituency seats. The problem they faced in 2015 was that the polling data turned out to be inaccurate, but if you feed the correct general election numbers then I think they could correctly call the result to within a very few seats (right now its showing an exact match but I suspect that they've retuned the model to allow for the 2015 result so I don't think that means the model is 100% accurate).
Some discussion here:
http://www.electoralcalculus.co.uk/trackrecord_15errors.html
It's obviously wrong to pretend that any prediction of the 2015 election was right, but I think to target Electoral Calculus in particular and call it "discredited" smacks of misunderstanding what that site even is. No surprises there.
In terms of the current prediction, the model is probably a little vulnerable to the impending boundary changes so that will add an extra error, but yes, again, there is no point in pretending that Labour are doing well. The boundary changes just mentioned are expected to hurt them anyway, losing around ten more seats than you might expect, so that doesn't help matters. On the other hand there are still three years to go so any prediction made now is unlikely to reflect the situation in three years' time on the other side of Brexit. I tend to expect that the current prediction is a bit in between what will happen after that. Theresa May will be a hero of the people, or a villain who failed to deliver anything like what was promised and satisfied exactly no-one, so I'm generally expecting that there will be a huge victory off the scale of this prediction, or a crushing Tory collapse that leaves voters with little viable option but to turn to Labour. Which of those becomes more likely we'll have to wait and see.
But anyway: Electoral Calculus did rather well in some respects in 2015, but on the usual computer modelling rule of "Garbage in, Garbage out", suffered from errors in polling giving it the wrong data to feed on.