Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMay 22, 2026Vermont, USA

SpecShift SPS Evaluator

Benjamin J. Clark, Oluwagbemike Olowe · Team SpecShift Research

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

Read the report

Report: SpecShift SPS Evaluator

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

SpecShift SPS Evaluator is a public synthetic scaffold for observable-only review of generated-code tasks.

The project explores a simple problem: passing visible tests is not the same as satisfying the underlying specification.

The prototype implements multiple structured review passes across baseline, schema, specification-perturbation, and adversarial conditions using a small synthetic dataset designed for local inspection and reproducibility review.

The system does not execute generated code or require model weights, private prompts, training data, or internal activations. Instead, it operates only on observable records and reviewer-visible artifacts.

The project is intended as a bounded research prototype for specification validation and reviewer-assisted discrepancy analysis, not as a production system or automated verifier.

Included artifacts:

* runnable local demo * synthetic dataset * structured review passes * public review packet * explicit limitations and boundary documents

The prototype also explores reviewer-controlled hidden-label evaluation structures for future observable-only assessment workflows.

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 report states its goal too briefly and lacks a detailed methodology. While the project seems to evaluate code generation across multiple stats/audits, it does not explain what each one is.

  2. Very bare-bones in terms of the document and your findings. But the bigger issue is with the idea: reward hacking is indeed an issue, and models will indeed often implement code that passes tests but does not fully implement the spec. But your proposal for fixing it is essentially equivalent to applying more test-time compute: which is already possible with claude code or codex (using /goal or dynamic workflows, for example). As models get better and more test-time compute is applied, basic types of reward hacking will get fixed by default, to be replaced with higher levels of abstraction of reward hacking. So a good way to intervene on this problem long term is to figure out how to do spec-driven development at the workflow level, in a way which can reduce these failure modes.

Cite this project

@misc{clark2026specshift,
  title = {{SpecShift SPS Evaluator}},
  author = {Benjamin J. Clark and Oluwagbemike Olowe},
  year = {2026},
  month = may,
  note = {Submitted to The Secure Program Synthesis Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/specshift-sps-evaluator-avwo}},
  url = {https://apartresearch.com/sprints/projects/specshift-sps-evaluator-avwo}
}

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