Directional Reasoning Trajectory Change (DRTC): Identifying Critical Trace Segments in Reasoning Models
This work addresses the problem of interpretability for researchers and practitioners by providing a causally grounded method to understand reasoning dynamics in language models, though it is incremental as it builds on existing interpretability approaches.
The paper tackles the challenge of interpreting long-horizon reasoning in language models by introducing Directional Reasoning Trajectory Change (DRTC), a process-causal framework that identifies critical decision points and measures how earlier context steers reasoning, achieving a median delta of 0.409 in a scaling study on 500 MATH problems.
Understanding how language models carry out long-horizon reasoning remains an open challenge. Existing interpretability methods often highlight tokens or spans correlated with an answer, but they rarely reveal where the model makes consequential reasoning turns, which earlier context causally triggers those turns, or whether the highlighted text actually steers the reasoning process. We introduce Directional Reasoning Trajectory Change (DRTC), a process-causal framework for interpreting long-form reasoning from a single on-policy rollout. DRTC detects pivot decision points using uncertainty and distribution-shift signals, then applies receiver-side interventions that preserve the realized rollout without resampling the continuation while blocking information flow from selected earlier chunks only at a pivot. It measures whether each intervention redirects the direction of the model's log-probability trajectory relative to the realized rollout direction, producing a signed per-chunk attribution score. We also compute turning-angle curvature changes on raw logits as a complementary diagnostic and introduce curvature signatures to summarize shared intervention-response geometry. Empirically, directional influence is sharply concentrated across four reasoning models (per-example |DRTC| shares yield Gini 0.50 to 0.58 and top-5 percent mass 0.23 to 0.28), and learned pivots induce stronger intervention magnitudes than matched random spans. In a scaling study on 500 MATH problems with R1-Distill-Qwen-1.5B, learned spans outperform matched random spans (median delta = 0.409, 355 of 500 positive; sign test p = 2.3e-21). Overall, DRTC provides a causally grounded, trajectory-level view of how specific context elements steer reasoning under on-policy dynamics.