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

Three Checks Before Calling It Collusion: Competence, Chance and Incentive Audits of LLM Pricing Agents

Demchyna Oleh

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

Read the report

Report: Three Checks Before Calling It Collusion: Competence, Chance and Incentive Audits of LLM Pricing Agents

Open in a new tab (opens in new tab)
Presentation

Presentation: Three Checks Before Calling It Collusion: Competence, Chance and Incentive Audits of LLM Pricing Agents

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

We audited our own LLM-collusion benchmark and found it measured noise. A local Qwen3.5-9B passed a table-based qualification gate (12/12) yet chose strictly dominated actions at the uniform-random rate (16/20 vs 15.75); adding a payoff table cut errors to 6/20 (McNemar p = 0.002). In 21 preregistered Bertrand-duopoly games with a 70% collusion premium (15 labeled-table, 3 matrix-table, 3 formula-only), all 15 labeled-table games priced inside or below the uniform-random band; talk did not raise prices, and agents honoured their "both set 21" proposals in 10/49 cases. The matrix table put 3/3 games above that band (exact p = 0.012; mean price ≈ 19 vs Nash 15). On scripted ground truth, level-based screens flag high prices, not strategy; only an end-game test separates strategic collusion from unconditional cooperation. We provide a three-check audit protocol and raw token journals.

Reviews

Judging this Sprint?

Review “Three Checks Before Calling It Collusion: Competence, Chance and Incentive Audits of LLM Pricing Agents”

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. The three checks framing in this study is genuinely useful. Before calling anything collusion, check that the agent is competent in the game’s own format, that the signal beats a uniform-random null, and that the environment actually rewards collusion. The second thing I liked is the detector calibration - level screens and collusive-message counts can catch high prices but cannot distinguish strategic collusion from plain cooperation, which only the end-game test can firmly establish. The one thing I would like to see in an extension of this study is the format effect replicated with more games and a second model so it can be settled rather than suggestive.

  2. Before calling something collusion, check that the agent understands the game, beats random play, and is actually rewarded for colluding. Their own benchmark failed all three, and a random agent looks "collusive" 99% of the time, which is a useful warning for the field. But 400 of 401 reasoning calls were truncated, the table-format effect rests on 3 games and didn't replicate, and the preregistration was only stored locally.

Cite this project

@misc{oleh2026three,
  title = {{Three Checks Before Calling It Collusion: Competence, Chance and Incentive Audits of LLM Pricing Agents}},
  author = {Demchyna Oleh},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/three-checks-before-calling-it-collusion-competence-chance-and-incentive-audits-of-llm-pricing-agents-n1z9}},
  url = {https://apartresearch.com/sprints/projects/three-checks-before-calling-it-collusion-competence-chance-and-incentive-audits-of-llm-pricing-agents-n1z9}
}

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