Why PeerPanel exists.

The first time I watched a colleague get desk-rejected on a fixable issue (a missing baseline comparison, two months of work) I thought there had to be a better way to catch this before submission. Not better peer review. Earlier peer review. The kind that happens while you can still revise.

I tried the existing AI review tools. They were a single model writing five paragraphs of generic feedback. They could not catch what a real specialist reviewer catches, because they were not specialists. They could not argue with themselves. They produced output that sounded thoughtful and was, on inspection, mostly surface.

So I built the architectural alternative. Five domain experts that review the paper independently, then are forced to challenge each other with quoted paper evidence. An editor that re-reads the paper to weigh every finding. Zero data retention, no author metadata in prompts, transparent scoring. The shape of this system is not original. It is grounded in published research on multi-agent debate and adversarial verification, but the application to pre-submission review is, as far as I can tell, new.

PeerPanel is in research preview right now. We are learning from real reviews from real researchers. The methodology page is being written. The paid launch is in July. If you find a bug or a missed finding, write to me directly. That is the most useful thing you can do during preview.