Skip to content
Apart Research
Apart Research
Contact us
Sprint projectFeb 2, 2026Newark USA

The Sentinel Engine: Solving the Observability Trilemma via Differential Precision Probing

Ramakrishna Doradla Venkatesh, Bryan Chan · Team Sentinel

Submitted to The Technical AI Governance Challenge. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: The Sentinel Engine: Solving the Observability Trilemma via Differential Precision Probing

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

The Sentinel Engine is a novel forensic framework that resolves the Observability Trilemma—the fundamental conflict between Inference Efficiency, Cognitive Observability, and Adversarial Assurance in LLM governance. By shifting the paradigm of quantization noise from a computational artifact to a differential structural probe, the system detects sophisticated "Silent Hackers" and sleeper agents in real-time

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. This is going after one of the most important problems in AI safety: deception / sleeper agents. The core idea is genuinely interesting: use sensitivity to precision / quantization noise as an audit signal for “deception-like” internal states. The theory is actually thoughtful and non-trivial. This is not a generic “train a probe” project.

    The weak point is the evidence. The results look too clean for a problem this hard, which makes me worry about confounds (like task difficulty / uncertainty) or evaluation artifacts. To make this convincing, I want: (1) difficulty-matched non-deceptive controls, (2) generalization across models + tasks, and (3) a real adaptive adversary where the model tries to hide from the detector.

    On presentation: I’m not penalizing theory. The theory is good. But I want the core claim + core experiment + key plots in a 1-page “here’s what we proved / here’s what we didn’t” summary. Make it impossible to miss what’s real vs what’s a hypothesis.

    Overall: high-upside direction, but needs harder validation.

    Read full reviewShow less

  2. The core idea is a cool twist on standard probing. The main problem is that the paper doesn't spend enough time thinking about confounders, e.g. the method may simply be detecting cognitive difficulty rather than deception. I'm also very skeptical of the AUC=1.0 result, and that merits more investigation. My recommendation is to spend more time thinking about how to interpret your results, and treat outlier results with skepticism.

Cite this project

@misc{venkatesh2026sentinel,
  title = {{The Sentinel Engine: Solving the Observability Trilemma via Differential Precision Probing}},
  author = {Ramakrishna Doradla Venkatesh and Bryan Chan},
  year = {2026},
  month = feb,
  note = {Submitted to The Technical AI Governance Challenge, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/the-sentinel-engine-solving-the-observability-trilemma-via-differential-precision-probing-noqd}},
  url = {https://apartresearch.com/sprints/projects/the-sentinel-engine-solving-the-observability-trilemma-via-differential-precision-probing-noqd}
}

More from The Technical AI Governance Challenge

All 47 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026