Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 25, 2026Bellevue, Washington

Rotating verifier partners in DeepSeek-V4-Flash: an inconclusive 8-point drop in acceptance

Venkata Anirudh Devireddy

Submitted to AI Collusion Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Rotating verifier partners in DeepSeek-V4-Flash: an inconclusive 8-point drop in acceptance

Open in a new tab (opens in new tab)
More on drive.google.com (opens in new tab)
Share

In Shi, Zhang and Yang (2026), fixed pairs of LLM agents end up accepting each other's work when they verify each other, even though REJECT is the only compliant action. For DeepSeek-V4-Flash, I test how much of this depends on meeting the same partner every episode. I keep the original environment, vary only the partners, and run three conditions: ten fixed partner pairs; the same agents rotated so no pair meets more than once; and a scripted partner that always rejects. I predicted that rotating partners would change acceptance by less than 15 points. Fixed pairs (partners) accept 81.0% of the time and rotated pairs (strangers) 73.0%. The difference is +8.0 points (nominal 95% CI +1.0 to +16.0, 10 task slots). This meets neither pre-registered decision rule, so it is inconclusive. Even against a scripted partner that always rejects, the model accepts 63.0% of the time. Mutual ACCEPT between partners is 76.0%, compared with 83.0% in the released data, which clears the pre-set replication gate. The intervals are descriptive: agents in the rotating population have linked interaction histories, so its slots are not independent.

Reviews

Judging this Sprint?

Review “Rotating verifier partners in DeepSeek-V4-Flash: an inconclusive 8-point drop in acceptance”

Your public critique appears on the project page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on the project page.

Shown on the project page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; the project page shows only that a judge declined, not why.

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

See our Privacy Policy.

  1. This submission examines the effect of partner rotation on the frequency of collusion.

    Unfortunately, due to compute limitations, it is unable to draw statistically significant conclusions. I'd be curious to repeat this experiment with more samples and also with newer/larger models.

    Separately, I'd be interested in understanding what drives agent decisions to collude or not, how aware they are of partner rotation, etc.

Cite this project

@misc{devireddy2026rotating,
  title = {{Rotating verifier partners in DeepSeek-V4-Flash: an inconclusive 8-point drop in acceptance}},
  author = {Venkata Anirudh Devireddy},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/rotating-verifier-partners-in-deepseekv4flash-an-inconclusive-8point-drop-in-acceptance-a5su}},
  url = {https://apartresearch.com/sprints/projects/rotating-verifier-partners-in-deepseekv4flash-an-inconclusive-8point-drop-in-acceptance-a5su}
}

More from AI Collusion Research Sprint

All 392 entriesBrowse all projects

A Sprint is running now

AI x Epistemics Research Sprint · Nov 13 - 15, 2026