Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026Winnipeg, Manitoba, Canada

Causal-Interp: Auditing Counterfactual Sensitivity in Mechanistic Interpretability

Ranveer Gill · Team Causal-Interp

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

Read the report

Report: Causal-Interp: Auditing Counterfactual Sensitivity in Mechanistic Interpretability

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

Causal-Interp is an open-source pipeline for auditing whether mechanistic interpretability findings remain stable under changes to the counterfactual intervention used to produce them. Across three published transformer circuits, I test activation patching, counterfactual disagreement, calibration, replication, and resample ablation. The clearest result is that, on a published docstring circuit, changing only the counterfactual while keeping the model, clean prompts, metric, tested positions, and cutoff fixed changes recovery from 3/6 to 5/6 published heads. Further experiments show that reproducibility and high circuit-level recovery do not by themselves establish that a causal explanation is complete or trustworthy. The project provides an auditable tool and evidence standard for stress-testing mechanistic claims that could inform AI incident investigations.

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. Careful and well executed, with unusually clear evidence boundaries. However, the main finding—that patching results depend on the counterfactual and ablation design—is already established in prior work. The experiments are limited to three small benchmark circuits, so the claimed relevance to real AI incident response remains speculative.

Cite this project

@misc{gill2026causalinterp,
  title = {{Causal-Interp: Auditing Counterfactual Sensitivity in Mechanistic Interpretability}},
  author = {Ranveer Gill},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/causalinterp-auditing-counterfactual-sensitivity-in-mechanistic-interpretability-ouk7}},
  url = {https://apartresearch.com/sprints/projects/causalinterp-auditing-counterfactual-sensitivity-in-mechanistic-interpretability-ouk7}
}

More from AI Incident Response Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026