REF final guidance arrives

The final guidance on Research Excellence Framework submissions, and panel criteria and working methods, has now been confirmed ahead of the November 2028 deadline. Like many of the updates to the REF this is an evolution of work – but the release of the final guidance crystallises some of the measures on quality, impact, and environment.

To briefly recap how we got here: REF 2029 differs from its predecessors in that it has a much stronger emphasis on the conditions in which research is done, not just the quality of the final output. To achieve this ambition the REF team initially proposed increasing the weighting of people, culture and environment by ten per cent, removing the cap on the maximum amount of outputs from one individual, and tightly controlling the portability of outputs.

In sum, it was hoped that this approach would mean that REF more closely reflected the aggregate research performance of an institution rather than simply adding up the performance of individuals as in previous iterations.

Over the last few years, there have been fierce debates on whether this approach could work. There have been criticisms that this is a devaluation of impact, it is too complicated, too restrictive around outputs and portability, too inflexible on open access, and open to gaming around equality measures. The debate reached a fever pitch when then science minister Patrick Vallance announced a pause in September 2025 until December 2025.

It is never a good sign when a minister is taking an interest in the ins and outs of units of assessment and data definitions. Come December 2025 the REF team made significant revisions, including replacing People, Culture and Environment (PCE) with Strategy, People and Research Environment (SPRE). This reduced the overall weighting of the measure to 20 per cent while placing a greater emphasis on research environment, with the term “culture” downplayed. Assessment of Contributions to Knowledge and Understanding (CKU), the old outputs measurement, was also simplified to look more like REF 2021.

It’s worth pausing to recognise that while the REF team has dialled down some of its initial decisions, the overall shape of the new exercise is still fundamentally different. In the final set of guidance, confirmed today, environment measures will still contribute 20 per cent to the overall rating, volume measures are based on average headcount and not a census, the cap of five maximum outputs from any one individual can be waived (though this will require explanation), and the rules around open access remain more far-reaching than other versions.

The latest update

In today’s final guidance there is no sign that the REF team are dialling down their ambitions for a REF aimed at “team science.” As expected, many of the updates are at the far end of technicality and aimed at the unfortunate souls preparing the gigantic submissions from within their institutions. This merry few may be pleased to learn that, for the first time, the four main panels have agreed shared criteria and quality definitions. Ideally, this should lead to greater consistency across disciplines with the addition of supplementary guidance developed for each panel.

The most substantive announcement, and likely the one provoking most immediate scrutiny, is around the final shape of SPRE: a 10,000 word institutional-level statement (2,000 words for context, 8,000 used flexibly over four scored sections) and another 10,000 word unit-level statement.

This latter document will contain a Statement on the Research Community – explaining the range of roles, functions and career stages in the unit – and a Statement of Representation, to be used in understanding how representative the REF submission has been. One of the major risks with the new REF is that with a much looser cap on the maximum number of outputs institutions could put forward their star academics to the detriment of others. To mitigate against this, units will also have to explain how their submission is representative of the unit as a whole.

This establishes a link between the explanation of the diversity of a submission and the scoring of its outputs. Research environment should not therefore be able to exist separately from outputs within a scoring mechanism that penalises submissions that are not reflective of a range of research activity. As the new panel guidance puts it:

“the extent to which there is coherence between the breadth of outputs submitted for CKU and the evidence of support for breadth of research activity in the scored sections of the ULS [unit-level statement] will be used to inform the sub-panel’s assessment of the ULS.”

How this works in practice is no small job for the panels to get to grips with – and is one of the many areas where those compiling and overseeing submissions are likely to feel somewhat tentative.

The same may be true for much of the twin SPRE sections, though there is now detailed guidance and suggested – though only suggested – indicators and metrics that could be used to support the long, complex narratives.

So institutions are “strongly advised” to use longitudinal data to evidence outcomes related to research culture and environment enhancement, for example, but there are many possibilities to choose from: staff experience, pay gaps, progression, induction, possibly even data on bullying, whistleblowing, and research misconduct. Equally, to evidence research strategy, narratives could be supported by indicators such as evidence related to concordats and charters – and there are a plethora to pick from.

The panel guidance also provides advice on use of AI. Panel members will not be allowed to upload content to an AI tool, nor allowed to use AI to generate judgements on quality, but they may “make use of AI/automation to process scoring data, with a view to improving the robustness of assessment – for example, by correcting for systematic differences in scoring behaviour of panel members.” There is also scope for AI to be used for “strictly administrative purposes” by the REF team. This is a very tentative step toward increased technology use – it will be interesting to see the reported efficiency gains of using AI, and crucially, how Research England and its partners will monitor its use.

There’s still plenty left to say about the final shape of the exercise and lots of detail to unpick, not least how some of the ambitions mesh with a sector in a gathering financial crisis. But after years of debate the REF is now entering its home stretch.

It felt for a while it might not get this far, and yet here we are approaching another submission deadline. The exercise which always seems too expensive and too complicated is at the apex of another cycle. It will be another few years until we know whether it was all worth it.