Computer science conferences are in crisis because of a glut of submissions that overwhelm reviewers. ICSE 2026 had ~1500 submissions, up from about ~800 in 2023, and other conferences are similar. Too many of these submissions have low quality. (The ICLR 2026 chairs reported "possible cases where authors submitted multiple slightly different variants of the same paper (LLM-paraphrased or otherwise)".) Such submissions are easy to prepare and submit, but reviewing them is burdensome.
To limit how much reviewing one submitter can demand, and to restore the incentive to submit only high-quality work, I propose to limit the number of submissions to a conference. Let N be the maximum number of papers you have ever had accepted to a given track (such as the research track) at a single edition of conference C. Then you may submit at most N+3 papers to that track of conference C this year. Each submission counts against the limit of every one of its submitters. A review system such as HotCRP can refuse to register a paper that would put any of its submitters over the limit.
As an example, if you had 1 paper accepted to the ICSE 2021 research track and 3 to the ICSE 2023 research track, then N=3 and you may submit 6 papers to the ICSE 2027 research track. However, a submitter who is new to ICSE can appear on at most 3 papers.
This rule improves incentives. Submitters must choose their best papers, so the slop gets left out, instead of forcing reviewers to filter it out at great cost.
This rule does not unduly constrain newcomers, such as new grad students, who have N=0 and may submit 3 papers. Such individuals are unlikely to produce more than 3 high-quality submissions to one track of one conference in a year.
The rule does not unduly constrain veteran submitters with an established track record, who can submit more (and are likely to be active reviewers in the community).
The rule mostly affects rejected submissions, which is where the slop is concentrated.
Here is data that supports the proposal:
- For ICSE 2014–2025, the research track had 6963 author-years (an author who published in 5 of those years counts 5 times). In only 8 out of 6963 author-years did an author have more papers accepted than N+3 allowed. Dropping 13 papers out of the 1715 published would satisfy N+3 for all authors in all years.
- For ICSE 2026, there were 1473 submissions. More than 1/3 of the submissions were authored by someone who violated the N+3 rule. Dropping 171 of the 1473 submissions would satisfy the rule.
In other words:
- The cost is very low, for authors and the community: 0.8% of accepted papers would be rejected (for ICSE 2014-2025).
- The benefit to the community is a reduction of 12% of reviewing load (for ICSE 2026).
Possible objections:
- Does this discourage collaboration? Probably not; if so, only at the margin, and the data show that the margin is small. It does discourage gift authorship (co-authors who did minimal work), and it discourages repeatedly re-submitting low-quality papers.
- Why not use a sliding window to compute N? For simplicity. Also, successful researchers are unlikely to turn into slop factories.
- Does this favor incumbents? The rule permits growth: if all your N+3 papers are accepted, then next year's limit is N+6.
- What about large labs, such as a PI with 15 students but N=2? The data show that this occurs rarely. The rule incentivizes the PI to focus on quality over quantity, at least for conferences that use the N+3 rule.
- Why not use reviewing quotas (mandatory reciprocal reviewing, as NeurIPS and ICLR use)? It is frequently criticized for producing low-quality reviews.
- Why not use some other similar formula? That could work; the exact constants are not critical. Some conferences use flat caps; by contrast, the N+3 rule takes account of demonstrated track record and reduces the slop more than a flat cap does.
I have been proposing the N+3 rule for many years — since before LLMs, but after some submitters began flooding conferences with low-quality papers. Now is the time for conferences to adopt the N+3 rule. I encourage steering committees and program chairs to do so.
