Horizon Europe evaluation criteria: how proposals are actually scored

There is a version of grant writing where you tell the story of your project as well as you can and hope it lands. There is another version where you know that three named criteria will each be scored and each carry a minimum, that one of them may weigh more in the ranking, and that a single weak criterion ends the process, and you write accordingly. The second version wins more often, and it is not because those applicants are better writers.

The three criteria and what each one asks

Horizon Europe research and innovation proposals are evaluated against Excellence, Impact, and Quality and efficiency of the implementation. The wording is adapted for some action types and for the European Research Council and Marie Skłodowska-Curie Actions, which run their own schemes, but the underlying structure holds across most collaborative calls.

Read those middle-column questions again. Every one of them asks for evidence, not intention. That is the single most useful thing to understand about EU evaluation: the panel is not judging whether your project is worthwhile. It is judging whether your document proves it.

The scale, and what actually separates the grades

Each criterion is scored on a short scale, half marks are allowed, and the scale is defined in terms of how well the proposal addresses the criterion, not how good the project is. The published wording moves from a proposal that fails to address the criterion, through poor, fair, good and very good, up to excellent.

Thresholds, weighting and the ranking

Three things decide where a proposal ends up, and they are not the same thing.

1. A minimum on each individual criterion. A criterion below it removes the proposal from consideration, no matter how strong the other two are.
2. A minimum on the total. The sum of the criterion scores has to reach it.
3. Weighting, which applies to the ranking only. Scores are normally not weighted at all. Where a weighting does apply, as it does to the Impact criterion for Innovation Actions, it is used to determine the ranking and plays no part in deciding whether you passed the thresholds.

Two-stage calls work differently again, evaluating only part of the criteria at the first stage and applying their own thresholds.

Before you plan any work, read the call conditions and the General Annexes of the work programme that governs your topic, and take the applicable values from there. Specific calls and topics may set different rules on thresholds and weighting, and the binding numbers are the ones in your own call.

Source: the Horizon Europe Work Programme 2026-2027 General Annexes and the European Commission's standard evaluation form for Research and Innovation Actions and Innovation Actions.

The practical consequence is unglamorous and important. Passing the thresholds is only the entry ticket, and funding is decided by the ranking. Where a criterion carries extra weight in the ranking for your action type, a point recovered there is worth more than the same point recovered elsewhere, and your remaining time should go there first. Most teams instead spend the final days on whichever section they find most interesting to write.

What evaluators write about you

Scores do not travel alone. Each criterion comes with recorded strengths and weaknesses, and after the consensus discussion these become the Evaluation Summary Report that applicants receive. Reading past reports, from your own applications or from colleagues, is the cheapest training available in this field.

The language is consistent and worth learning to hear in advance:

• "The objectives are only partially quantified" means no numbers where numbers were expected.
• "The pathway to impact is not sufficiently substantiated" means you asserted the outcome and skipped the mechanism.
• "The role of some partners remains unclear" means a partner appears in the consortium table and nowhere in the work plan.
• "The risk analysis is generic" means the risks could belong to any project in the programme.

Each of those sentences costs between half a point and a full point. Three of them in one criterion is the difference between very good and good, which in a competitive call is the difference between funded and not.

Why a good score is still not funding

Passing the thresholds makes a proposal eligible for funding, not funded. Proposals above threshold are ranked, and the call budget is allocated down the ranked list until it is exhausted. In oversubscribed calls, strong proposals are routinely left above threshold and below the funding line. Where proposals tie on score, the General Annexes set out a defined order for separating them, starting with proposals that cover aspects of the call other better-ranked proposals do not, and then working through the individual criteria and further factors in a stated sequence.

This is why marginal points matter more in EU funding than in almost any other kind of writing. You are not trying to be good. You are trying to be ahead of the proposals immediately around you, and those are also good.

How to self-score before you submit

Run this exercise at least a week before the deadline, while there is still time to act on what it tells you.

1. Rebuild the scorecard from the call. Write out the criteria, sub-aspects, weights and thresholds exactly as your call states them. Not from memory, and not from a previous call.
2. Map each sub-aspect to a location in your document. Page and section. Any sub-aspect with no location is a shortcoming waiting to be written down.
3. Score each criterion independently, in writing. Force yourself to name the shortcomings that keep it off 5. If you cannot name any and you gave a 4, you did not score honestly.
4. Have someone who did not write it score it too. Give them the criteria and nothing else, no context and no explanation. Their score is closer to the truth than yours.
5. Fix in weighted order. Points recoverable, multiplied by weight, divided by effort. Work down that list, not down the document.

Our pre-submission checklist covers the compliance layer that sits underneath all of this, and how evaluators actually read proposals explains the conditions the panel works under.

Where general AI falls short on this specific job

The self-scoring exercise above is exactly the sort of task people hand to a chat assistant, and it is exactly where a chat assistant is weakest. A general model has no fixed record of your criteria, applies no weights, does not hold a threshold in mind, and will produce a different score for the same document on a second run. It cannot tell you that a criterion sits at 2.5 and therefore ends the application, because it never encoded the threshold in the first place. That is explained in more detail in what ChatGPT, Claude and Gemini miss on grant proposals.

The Evalora Engine was built around this exact scorecard. It reads your official call document and extracts the criteria, weights, eligibility rules and required sections. It scores your draft criterion by criterion against that structure, marks unmet requirements, ranks improvements by the points they recover, and runs an evaluator simulation that returns a predicted score with reviewer-style comments per criterion. Every run is saved, so you can see the score move as you fix things rather than hoping it did.

Frequently asked questions