🤖 AI Summary
This work addresses the lack of auditable reasoning mechanisms in large language models, which rely solely on statistical correlations and thus struggle to support trustworthy agent decision-making. To bridge this gap, the authors propose TRACE, a novel framework that integrates formal reasoning theory into a typed, versioned structure for recording reasoning trajectories. By introducing a record-consumer contract mechanism, TRACE transforms reasoning traces into actionable interfaces. The framework features a causally specialized TraceRecord structure, an eight-stage reference writer, and a gated priority metric, complemented by the TRACE-Bench protocol and counterfactual regret analysis. Experiments in music instruction and flood rescue scenarios demonstrate TRACE’s ability to effectively disentangle associative, interventional, and normative reasoning, thereby establishing a comprehensive, auditable decision-support architecture.
📝 Abstract
This paper defines TRACE (Typed Reasoning And Commitment Evidence): a typed, versioned schema for recording reasoning traces, a reference procedure for writing records against it, and one operating discipline, no durable state change without a record. The paper argues in three layers that reasoning is not in the language model: the autoregressive mechanism natively computes association; chain-of-thought and reinforcement learning inherit its limits; and the formal constructs of reasoning theory, from Socratic procedure to Pearl's ladder, are absent as machinery. The schema answers the absence with fields and tests: the TraceRecord and its causal specialization, an eight-stage reference writer, a gate-first measurement regime, the TRACE-Bench protocol, and the consumers, memory admission, plan gating, temporal regret, and verdict reuse, whose more auditable decisions are the measure of the record. A record-consumer contract states what a record guarantees and what a consumer must honor in return, making the schema an operational interface rather than a passive document. Two worked examples run in the main text: a music-lessons argument traced from sentence to typed verdict, separating association, intervention, and prescription; and a flood search-and-rescue vignette in which a predictive world model reports confident plan success that its own support and out-of-distribution scores contradict, so the record defers the commitment, requests a bounded observation, revises append-only, and clears a different branch. The vignette is illustrative, not empirical; closed-loop evaluation is left to future work, so the contribution is the schema and its contract, not a performance claim. Appendices carry the full schema, writer algorithms and cost model, clinical and policy illustrations, the benchmark protocol, convergence metrics, and usage scenarios.