Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMay 23, 2026Hattiesburg

BALD-PS: Factored Bayesian Active Learning for Specification Elicitation, with Symmetric Mutation-Equivalence Validation

Barsat Khadka · Team OpenScience

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

Read the report

Report: BALD-PS: Factored Bayesian Active Learning for Specification Elicitation, with Symmetric Mutation-Equivalence Validation

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

BALD-PS is a framework for interactive formal specification elicitation from ambiguous natural-language requirements using factored Bayesian active learning. The key insight is that a specification naturally decomposes into two predicates , a precondition (requires) describing valid inputs and a postcondition (ensures) describing acceptable outputs , allowing disagreement among candidate specifications to factorize into independent axes. The system samples diverse specifications from multiple LLM personas, identifies high-information disagreements using a factored BALD objective, and queries an oracle/developer with maximally informative edge cases to efficiently resolve ambiguity. Unlike traditional version-space elimination, BALD-PS synthesizes new specifications by recombining surviving precondition and postcondition fragments, enabling recovery of correct specifications that no single agent originally proposed. To validate specification quality, the framework introduces symmetric mutation-equivalence tightness, a bidirectional mutation-testing criterion that checks both whether implementations refute mutated specifications and whether specifications reject mutated implementations. Empirically, the method achieves perfect recovery on a controlled benchmark and substantially outperforms interactive refinement baselines on live HumanEval-style tasks while requiring fewer oracle queries.

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. I have to admit, I find the presentation confusing enough that I'm not sure exactly what new approach is being suggested in the paper. The statistical terminology really throws me off, even as it *feels* like the idea here shouldn't require much prose to explain. The idea seems to be about finding illuminating questions to ask, in resolving spec ambiguities surfaced by candidate translations of the same natural-language spec; then using the results to come up with new, better translations.

    If this particular insight wasn't explored before, then a solid empirical evaluation could be very valuable. However, the paper contains no description of how the benchmarks were chosen, making it impossible to draw any real conclusions from the voluminous quantitative results and associated statistical terminology. All examples may have been cherry-picked to make this particular implementation look good.

    Read full reviewShow less

  2. Symmetric tightness is powerful.

    I am curious to know how to extend to k axes and how to determine the value of k.

Cite this project

@misc{khadka2026baldps,
  title = {{BALD-PS: Factored Bayesian Active Learning for Specification Elicitation, with Symmetric Mutation-Equivalence Validation}},
  author = {Barsat Khadka},
  year = {2026},
  month = may,
  note = {Submitted to The Secure Program Synthesis Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/baldps-factored-bayesian-active-learning-for-specification-elicitation-with-symmetric-mutationequivalence-validation-0jms}},
  url = {https://apartresearch.com/sprints/projects/baldps-factored-bayesian-active-learning-for-specification-elicitation-with-symmetric-mutationequivalence-validation-0jms}
}

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