Methods · Circuits in Mechanistic Interpretability
Circuit Tracing (SAE- or CLT-Based Circuit Discovery)
Survey sections: §7.2 Circuit Discovery with Replacement Models
Summary Circuit tracing discovers circuits in an interpretable replacement model: the original model’s computation is approximated with sparse linear features, for example from cross-layer transcoders, and circuit discovery is then performed over those features instead of coarse-grained components such as attention heads and MLP blocks.
A complementary approach first constructs a more interpretable replacement model (Section 6.6.2) that approximates the original model’s computation using sparse linear features, and then performs circuit discovery within this replacement model.
Assumptions and requirements
Recent work has implemented this paradigm using cross-layer transcoders (CLTs; Section 6.6.2) (Ameisen et al., 2025; Merullo et al., 2025). In particular, Circuit Tracer (Ameisen et al., 2025) replaces the model’s MLPs with CLTs while retaining the underlying model’s attention and residual-stream computation. This yields a prompt-specific attribution graph whose principal internal nodes are sparse CLT features rather than entire MLP blocks or attention heads.
Intended applications
For a given prompt, Circuit Tracer produces a sparse attribution graph that describes active features and their interactions, making it well suited to query-level circuit discovery.
Using automated feature-interpretation pipelines (Section 12.6) to assign natural-language descriptions to these features, human inspectors can more easily understand the discovered mechanism.
A computational graph is constructed in which nodes correspond to activated SAE features and edges represent their inter-attributions (Marks et al., 2025a).
Limitations
Interpretability platforms such as Neuronpedia (Lin, 2023) also offer interactive implementations. However, the approach still has several limitations. Specifically, CLTs only approximate MLP computations, so the discovered circuit may only cover a part of the model’s underlying mechanism. Moreover, feature-level attribution graphs can have many nodes and edges, so we must prune, cluster, or otherwise condense them before they become understandable.
Depends on feature quality; requires additional computation to train and analyse SAEs/CLTs
Related methods
Used in safety applications
Key papers
Cited in §7.2.
- PreprintEmmanuel Ameisen, Jack Lindsey, Adam Pearce, Wes Gurnee, Nicholas L. Turner, Brian Chen, Craig Citro, David Abrahams, Shan Carter, Basil Hosmer, Jonathan Marcus, Michael Sklar, Adly Templeton, Trenton Bricken, Callum McDougall, Hoagy Cunningham, Thomas Henighan, Adam Jermyn, Andy Jones, Andrew Persic, Zhenyi Qi, T. Ben Thompson, Sam Zimmerman, Kelley Rivoire, Thomas Conerly, Chris Olah, and Joshua Batson. Circuit tracing: Revealing computational graphs in language models. Transformer Circuits Thread, 2025. URL https://transformer-circuits.pub/2025/attribution-graphs/methods.html.
- PreprintJack Merullo, Connor Watts, Max Loeffler, Liv Gorton, Elana Simon, Tom McGrath, and Owen Lewis. Replicating circuit tracing for a simple known mechanism. https://www.goodfire.ai/research/replicating-circuit-tracing-for-a-simple-mechanism, 2025.
- PreprintJohnny Lin. Neuronpedia: Interactive reference and tooling for analyzing neural networks, 2023. URL https://www.neuronpedia.org. Software available from neuronpedia.org.
Also discussed in: §7.1 Circuit Discovery Methods · §6.6.2 Replacement Models and Attribution Graphs · §7.5 A Unified Evaluation Framework for Mechanistic Interpretability · §13.3.2 Mechanistic Monitoring · §13.3.2 Stage 2 – Circuit Discovery · §13.7.4 The Scalability Bottleneck
Other methods in Circuits in Mechanistic Interpretability
Activation Patching · Path Patching · ACDC · Causal Scrubbing · Interchange Interventions