Trajectory-Aware Retrieval Agents for Temporal Decision- Making
For practitioners needing to make decisions from temporally structured text (e.g., medical, financial), TLM provides a method that preserves temporal context and yields consistent gains over existing approaches.
TLM introduces a closed-loop agentic framework that uses SHAP-guided feedback and a latent growth curve model over retrieved chunk embeddings to iteratively refine evidence for temporal decision-making. It outperforms zero-shot LLM baselines and standard RAG on medical QA, earnings call surprise prediction, and stock gap prediction tasks.
We study the problem of decision-making from long-form, temporally structured text using large language model (LLM) agents. Standard retrievalaugmented generation (RAG) pipelines fragment chronological context into isolated snippets, discarding the temporal structure that is often critical for correct downstream decisions. We introduce TLM (Trajectory Language Model), a closed-loop agentic framework that iteratively refines the evidence set using SHAP-guided feedback. The key technical contribution is the latent growth curve model (LGCM) over retrieved chunk embeddings, which provides an interpretable mechanism for detecting trajectory trends, turning points, and information gaps. We show that, under a scorer-calibration assumption (which holds approximately in practice), the iterative refinement procedure is monotonically non-decreasing in the probability assigned to the correct label. Empirically, TLM is evaluated on three temporally grounded decision tasks: medical question answering, earnings call surprise prediction, and overnight stock gap prediction. TLM substantially outperforms both zero-shot LLM baselines and standard retrieval-augmented approaches on the medical task, and yields consistent, economically meaningful gains on the two financial tasks.