Obviously margins of error apply (4.8% here), and nicer to have more sample than not - but past a certain point it’s quality that matters. It’s good to be sceptical but for the right reasons - if polling like this is wrong it’ll be because getting the right sample at constituency level is hard.