If you are working on an ERC proposal, our new app may help.

21 September 2026Zuzana HavránkováLinkedIn

It reads your draft against the ERC's own 2027 evaluation questions and flags problems. Each finding quotes your text and says what to change. It does not write the proposal for you. One prompt in a chatbot, nothing to install. Unofficial, and so far tested only on simulated proposals.

For applicants only: ERC rules bar evaluators from using AI on the proposals they assess.

If you use agents like Claude Code or Codex, the repository has an experimental deeper mode.

You do not have to open the app or the code at all. Just tell your AI assistant: check my ERC proposal using the tool by Irsova and Havranek.

Link in the first comment.

Links the author added in the comments:

https://tjhavranek.github.io/erc-ai-feedback/
https://github.com/tjhavranek/erc-ai-feedback
https://doi.org/10.5281/zenodo.20829165

Nothing is uploaded to us. You paste the prompt into your own chatbot, and the page has no analytics.

A square card headed 'ERC draft pre-review', marked unofficial. The headline reads: An unofficial check of your ERC draft before you submit. Under it: one prompt, pasted into a chatbot with your draft, read against the ERC's own 2027 evaluation questions, with every finding quoting your text and saying what to change. A panel headed 'One finding, on a synthetic proposal' gives an example rated HIGH, that the power claim contradicts the stated attrition assumption, quoting a sentence about 1,800 respondents per country against a 2,000 target, and asking for power to be recalculated on the retained sample. The footer gives tjhavranek.github.io/erc-ai-feedback, free, MIT licence, Tomas Havranek and Zuzana Irsova, Charles University.

Originally posted on LinkedIn. Archived in full among all posts.