Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMay 23, 2026Mt. Juliet, TN

When Models Disagree: Cross-Model Divergence Analysis for Ambiguity Risk Estimation in Software Requirements

Jack Lakkapragada · Team Jack

Submitted to The Secure Program Synthesis Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: When Models Disagree: Cross-Model Divergence Analysis for Ambiguity Risk Estimation in Software Requirements

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

Presentation: When Models Disagree: Cross-Model Divergence Analysis for Ambiguity Risk Estimation in Software Requirements

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

As AI systems generate increasing volumes of software code, one bottleneck in trustworthy software development is increasingly shifting upstream: specifying what code should do, and verifying that it does so. We investigate whether cross-model divergence in specification generation can serve as a practical signal for requirement underspecification. We present a structured experimental framework — a 30-requirement dataset spanning three ambiguity tiers, a six-category divergence taxonomy, and a normalized JSON comparison pipeline — applied across three contemporary LLMs: Claude Sonnet 4.6, GPT-4o, and Llama 3.3 70B. Divergence increases with pre-labeled ambiguity tier (TDR: 7.2, 7.4, 7.9), assumption divergence dominates across all tiers, and contradiction collapses to zero in adversarially ambiguous requirements. Operationally clear requirements still exhibit substantial latent interpretive variance, with many false consensus events in Tier A concentrated in security constraints. Framework, taxonomy, and dataset were frozen prior to execution, enabling reproducible analysis without post-hoc adjustment.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

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

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety 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 AI safety 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.

  1. The pre-registered protocol with frozen prompts and model blinding is a real strength, but the headline tier divergence rate of 7.2, 7.4, and 7.9 is too compressed by the ceiling effect to support much, and an FCR of 1.0 across every tier means that metric is not discriminating at all. The category distribution, especially the contradiction collapse to zero in Tier C, is the more convincing signal and would be a better thing to lead with. The shared training data across Claude, GPT-4o, and Llama also means convergence cannot be read as clarity, which caps how far the divergence-as-signal claim can go without ground truth.

  2. Research topic is interesting and important. Would be useful to extend to more test cases and also making the grading scale such that differences in capability are more clearly visible.

Cite this project

@misc{lakkapragada2026models,
  title = {{When Models Disagree: Cross-Model Divergence Analysis for Ambiguity Risk Estimation in Software Requirements}},
  author = {Jack Lakkapragada},
  year = {2026},
  month = may,
  note = {Submitted to The Secure Program Synthesis Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/when-models-disagree-crossmodel-divergence-analysis-for-ambiguity-risk-estimation-in-software-requirements-m8hx}},
  url = {https://apartresearch.com/sprints/projects/when-models-disagree-crossmodel-divergence-analysis-for-ambiguity-risk-estimation-in-software-requirements-m8hx}
}

More from The Secure Program Synthesis Hackathon

All 61 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026