The tension at the heart of peer review is not about expertise. It is about ego, taste, and the quiet resistance to changing one's mind. The anonymous reviewer who acknowledges that their concerns were addressed, then holds the score hostage because they "don't vibe" with the paper, is not practicing rigorous science. They are practicing a form of intellectual gatekeeping that prioritizes personal preference over the core purpose of review: to improve the work, not to validate the reviewer's worldview. This is a problem we should all care about, because it touches on how we handle uncertainty and bias in our own data pipelines. After all, if you are building models that clean data starts with catching AI slop before it skews your model, you already understand that noise corrupts outcomes. The reviewer's unadjusted score is noise in the scientific record.
This is not a small, procedural gripe. The post from the researcher, submitted to the NeurIPS 2026 discussion, cuts to a practical reality: when a rebuttal successfully addresses every listed concern, the score should move to reflect that reality. It should not remain frozen because the reviewer finds the methodology unappealing or the topic too niche. We understand the impulse to trust your gut, but the review process is not a referendum on whether you would have written the paper yourself. It is a check on whether the claims are sound and the evidence supports them. In the same way that exploring real-world computer vision forces you to accept that edge cases and deployment constraints matter more than your initial assumptions, peer review must force us to accept that a well-argued rebuttal changes the evaluation. If you do not adjust your score, you are telling the author that no amount of evidence will ever change your mind. That is not critical thinking; it is stubbornness.
Here is what we would tell any reviewer reading this: Your job is not to agree with the author's taste. Your job is to assess whether the work is valid, reproducible, and a contribution to the field. When you list concerns, you are making a contract. You are saying, "If you fix these, the paper is acceptable." Breaking that contract because you "don't vibe" with the approach is a breach of trust that erodes the entire ecosystem. It punishes authors for doing exactly what you asked. And it has a chilling effect on innovation. Why would anyone explore an unconventional idea, like the Forrester function as a tool for machine learning, if they know that a reviewer can dismiss it on a whim, not because the math is wrong, but because it feels unfamiliar? The beauty of scientific research is that we each get to explore ideas we find meaningful, even if their value is not immediately obvious to every individual reviewer. That is not a flaw to be managed; it is the point.
The concrete takeaway for authors is to demand a structured response in your rebuttal. Ask the reviewer to confirm, point by point, whether your responses are satisfactory. If they say yes, and they still keep the score, that is a failure of process, not your work. The open question for the community is whether venues like NeurIPS will codify this expectation into policy. We would argue they should. Because a review that refuses to move after its concerns are addressed is not a review. It is a veto, and it has no place in a system that claims to value the pursuit of knowledge over the comfort of the reviewer. Watch for that policy change. It will tell you more about the health of the field than any acceptance rate.