The most damning detail in this account isn't the adversarial reviews or the unresponsive area chair. It's the calibration gap. The author gave scores they considered fair, rejecting only for severe issues, while a reviewer who raised minor concerns slapped a reject with 1s across the board. That's not a difference of opinion; it's a fundamental breakdown in what the numbers mean. When reviewers and authors are operating on different scales of severity, the entire peer review process stops measuring quality and starts measuring luck. This isn't a new problem, but this framing makes it concrete: they tried to do the job responsibly, and the system rewarded that effort with a lottery ticket.
We see the same pattern in the Clean Data Starts With Catching AI Slop Before It Skews Your Model story, where automated detectors flagged genuine reviews and made the sentiment model less accurate. In both cases, the tool meant to filter noise ends up amplifying it. Here, the "AI slop" suspicion poisons the well before a single review is written. If an author suspects a paper is slop, they might review it more harshly, or worse, dismiss it outright. That's not peer review; it's pattern matching with a deadline. This experience suggests that even when you fight the urge to punish suspected slop, the system punishes you anyway.
What's striking is that this person did everything right. They engaged with the papers, they calibrated their scores honestly, and they only rejected for real issues. The result? Two adversarial reviews and an AC who vanished until the final hour. This isn't an outlier; it's the expected outcome of a process where accountability is optional. The Exploring Real-World Computer Vision: Deployments, Edge Models, and Current Challenges post highlights how practitioners build systems that must work in messy, real-world conditions. Peer review should be part of that ecosystem, but it's currently a black box where a single bad draw can sink a solid paper. The "lottery" framing is too generous. Lotteries are transparent about their odds; this process hides them.
The practical takeaway for anyone submitting to these conferences is grim but actionable: assume the review process is adversarial and plan your paper's defense accordingly. But that's a band-aid on a broken system. The real question is whether venue organizers will start auditing reviewer calibration, not just collecting scores. We'd tell a reader who asked that if your paper gets rejected, don't assume it's your work. But more importantly, don't let that discourage you from being one of the responsible reviewers the system needs. This attempt to do right is the only thing that kept this from being a total farce. The next step is forcing the system to meet that standard. Watch for whether the next NeurIPS introduces any form of reviewer feedback or calibration checks, because without that, the lottery gets worse every year.