Obviously these particular projects are always tricky and there are additional issues this time, but I do find myself pulling a face at some of the numbers there.
I think it's basically impossible to do this without some duff figures. Local election patterns are not always a good guide to General Election ones, especially where they involve the Lib Dems (and now I assume also the Greens) but sometimes with other parties as well. And demographic modelling, while useful in principle and perhaps able to help correct for the issues in the last sentence, isn't perfect either and can't cope with personal votes and targetting.
Whitehaven & Workington was discussed above, and his Con majority there seems a bit high. I'm also very surprised he has Spen Valley as close as he does (Con majority below 2000) and of course if that one is too Labour Dewsbury & Batley is going to be too Tory. Wolverhampton SE is actually known to have been narrowly Tory (ward breakdowns in both Walsall and Wolverhampton are in the public domain) and he has it as narrowly Labour while Wolverhampton W is similarly too Tory (it was actually very close).
Other examples?