circuit-discovery

Tracking exactly what happens to the internal "circuitry" (induction heads) of a 2-layer attention-only Transformer when forced to undergo domain adaptation from prose to structured Python code.

Circuit-level regression testing for AI systems. Catch mechanistic drift that behavioral evals miss.

Open-source EU AI Act Annex IV documentation toolkit. Mechanistic interpretability + circuit discovery for transformers. One function call generates a structured, hash-chained evidence package.

A framework for identifying neural circuit components via convergent evidence across weight deltas, activations, and latent geometry.

Circuit-level interpretability evidence does not survive defensible analytic variation. Pre-registered multiverse over 15,840 specifications; the derived EU AI Act Annex IV claim flips across 73.2% of specification pairs. arXiv:2608.13754

Official code of the project "Query Circuits: Explaining How Language Models Answer User Prompts" accepted to ICML 2026

Official implementation of "How Do Linear Probes Emerge? A Circuit-Tracing Framework with Concept-Targeted Attribution" (EMNLP 2026)

Practical mechanistic interpretability tools — activation caching, linear probes, activation patching, circuit discovery, and visualization for transformer models

Mechanistic interpretability CLI for transformer models on Apple Silicon. Analyze per-layer predictions, monitor activation drift, compare models, discover circuits. MLX-based, no GPU needed.

Mechanistic interpretability and safety auditing for Decision Transformers. Features circuit mapping, causal patching, TopK SAEs, and behavioral steering.