A researcher is sitting on a rebuttal that could define their submission, and the question is not about model performance or statistical significance. It is about whether to break a rule to make their point more clearly. Linking to figures in a NeurIPS rebuttal is worth the risk, given that the official guidelines say links are not allowed. The anxiety is palpable, and it is entirely misplaced. The real problem is not the researcher's judgment; it is that a venue as influential as NeurIPS still operates as if the internet does not exist.
The instinct to show results in a figure rather than a dense table is not just aesthetic. It is about respecting the reviewer's cognitive load. A well-designed plot can communicate in seconds what a table buries in rows of numbers. The fear that a link might trigger rejection, rather than a polite request to remove it, reveals a deeper issue: we have built a review culture that punishes initiative and rewards rigid compliance. This is the same tension that plays out when Clean Data Starts With Catching AI Slop Before It Skews Your Model highlights how automated filtering can reject valid signals because they do not fit a predefined pattern. The reviewer is not your enemy, but the system has made them a gatekeeper for rules that were written before modern collaboration tools were common.
Here is our honest take: the risk of a slap on the wrist is low, but the risk of a wasted opportunity is high if you let the format dictate the science. If the reviewers asked for additional experiments, they want to see the results. A link is a means to that end. The more practical approach is to embed the figure as a base64-encoded image in the markdown, which OpenReview supports, sidestepping the link rule entirely. This is not gaming the system; it is using the tools you have. The same logic applies to the broader conversation about Exploring Real-World Computer Vision: Deployments, Edge Models, and Current Challenges, where the gap between theoretical benchmarks and practical deployment forces you to adapt or fail. The rebuttal is a deployment problem, and your solution should be robust to the environment, not the other way around.
If you are weighing this decision, ask yourself what you would do if the rule were not there. Would you include the figure? Then you already know the answer. The real question is whether you trust the reviewers to respond to the substance of your work or whether you assume they will punish you for a minor formatting choice. In our experience, most reviewers are reasonable people who want to accept your paper if you give them a reason to. A clear figure that answers their question is a stronger reason than a link that they may or may not click. The Explore the Forrester Function: Beyond Mathematics, a Tool for Machine Learning post shows that even abstract mathematical tools become useful when you frame them in a way that connects to a practical problem. Your figure is no different.
The deeper issue is that OpenReview's markdown support is stuck in the past, and researchers are left to improvise. The person who posted this is right to call that out. But waiting for the platform to catch up is a losing strategy. Use the tools available, make your case clearly, and let the work speak. The consequence you should worry about is not rejection; it is submitting a rebuttal that buries your best evidence in a table and hoping the reviewer cares enough to dig. That is a risk no one should take. Watch for the response to your figures, not the rule. That will tell you more about your reviewers than any guideline ever will.