Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026Detroit

How much coalition advantage does one binary message buy?

Shashwat Bhatia · Team Decode

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

Read the report

Report: How much coalition advantage does one binary message buy?

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

Does a coalition of agents do better with one binary message per episode than with none? I added a one bit budget to the Colosseum Jira task and tested it on four models. A mediator lets one sender post one message, 0 or 1, in one fixed slot. The bit tells the partner which of two joint plans fits a hidden draw that only the sender can see. I compare one bit against a silent channel, paired by seed, with a shuffled bit as the content control and free talk as the positive control. The answer is inconclusive on every model I could judge. On Qwen2.5-7B-Instruct, over 30 paired games per arm, one bit beat silence by +0.49 points of coalition advantage (95 percent interval -3.67 to +4.98). On Qwen2.5-14B-Instruct-AWQ the gap was +0.81 (-5.36 to +6.38). The one clean result is on 14B, where the real bit beat the shuffled bit by +4.98 (+0.06 to +10.22). The 32B and GPT-5 mini batches failed the positive control, so they count as description only. The gate itself held in all 280 audited runs, and no arm harmed the other players.

Reviews

Judging this Sprint?

Review “How much coalition advantage does one binary message buy?”

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 paper is for track 3.1 but with a twist. Instead of asking which channels carry collusion, it asks how much channel is needed: one bit, enforced in code, on the Colosseum Jira task, against silence, with free talk as the positive control. This topic is chosen well and the paper is honest about its limits from the start.

    The really liked the execution and it is strong. The rules and the positive-control gate were frozen before the runs, and when the 32B and GPT-5 mini batches failed the gate, they were set aside as description-only rather than quietly kept. The shuffled-bit control is a nice way to separate what the bit says from the fact that a message arrived, and the gate held in all 280 audited runs. The repo reproduces: the numbers script confirms every headline interval. The primary comparison is inconclusive, and the paper tells it plainly. Presentation is well structured and easy to follow.

    My suggestions - the primary comparison is underpowered at 20–30 paired games, with per-seed swings of about ±45, so one strong plan-following model at 60 or more pairs would settle it, ideally once free talk clearly beats silence. Everything runs on one task and one coalition prompt, and the larger Qwen models are quantized while 7B is not, so a second task and matched precision would make the cross-model picture easier to read.

    Read full reviewShow less

Cite this project

@misc{bhatia2026much,
  title = {{How much coalition advantage does one binary message buy?}},
  author = {Shashwat Bhatia},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/how-much-coalition-advantage-does-one-binary-message-buy-fn6v}},
  url = {https://apartresearch.com/sprints/projects/how-much-coalition-advantage-does-one-binary-message-buy-fn6v}
}

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