Judge Circuits
For researchers using LLM-as-a-judge, this work provides a mechanistic explanation of format-induced inconsistency, showing that benchmark comparisons across formats may be confounded by formatting artifacts.
The paper identifies that LLM judges assign inconsistent scores across output formats (e.g., rating vs. True/False) due to a shared latent evaluator sub-graph in mid-to-late MLPs, which is then mapped through fragile, format-specific terminal branches. This implies that cross-format reliability comparisons partially measure formatter geometry rather than evaluation quality.
LLM-as-a-judge has become the dominant paradigm for grading model outputs at scale, yet the same model assigns systematically different scores when its output format changes (e.g., a 1-5 rating vs. a True/False label). Existing diagnoses of these format-induced inconsistencies stop at the input-output level. Using Position-aware Edge Attribution Patching (PEAP), we causally investigate the internal mechanism in Gemma-3, Qwen2.5, and Llama-3. We find that judgments across structured understanding and open-ended preference tasks share a sparse, generalized Latent Evaluator sub-graph in the mid-to-late multi-layer perceptrons (MLPs); zero-ablating it collapses judgment while preserving world knowledge in architecturally modular models. By structurally decoupling abstract judging from output formatting, we provide a mechanistic account of format-induced inconsistency on the open-weight models we study: a continuous judgment signal computed in the shared trunk is mapped through fragile, format-specific terminal branches, enabling format-independent preference to be isolated downstream of the requested output format. Our findings imply that benchmark-level reliability comparisons across formats are partially measuring formatter geometry rather than evaluation quality.