If universities want top research, they must allow people to fail

OK, but I thought that the REF only assessed whole units of assessment, so individual academics' scores for their outputs aren't published. In which case how does anyone have any idea what star rating they are aiming for? Frankly I neither know nor care whether my publications were rated 1 or 4, as I don't want my priorities to be determined by meaningless league tables, and don't want career advancement.

The REF reforms create space for failure by moving away from the assessment of individuals - Outputs and impact case studies are individual (or groups of individuals) products. The REF may not explicitly release the individual scores but they do assess each individual output. Internal REFs often determine progression and promotion. The most dangerous thing about the REF is that the individual scores are not released, which means no one can ever cross-check how the internal evaluations compare to the REF evaluation. This means not being given credit where credit is due (e.g. originality, significance and rigour are 4* but the paper appears in a 3* journal) or vice-versa. Individual REF scores must be released so that individuals can learn from the feedback and improve.

'Learn from the feedback and improve' = is what the journal refereeing process already does, done by actual experts. No need for the REF - still less internal fake REFs which are even more meaningless, especially as they sometimes ask you to rate your own work! However if you don't want promotion, internal REFs are easy to ignore and disengage from, by rating your own work as around 1 and getting back to actual research. We definitely don't need individual scores released for the real REF, because then there would be a risk of having to take the whole REF farce seriously.

If we are being evaluated we need to know how we fared. It is as simple as that. Whether such an evaluation itself should exist (ie should we have a REF) is a different question.