Divergent Paths: Learning an Outcome-Predictive Representation of LLM Agent Trajectories

Alexander Gill, Guram Gogia, Francisco Guzmán
10/14/2026
cs.LGcs.AIcs.CL

No official code found. Best implementation: alexgill321/precog-rm (0 stars, MIT, last commit 2026-10-14).

Abstract

In this work, we reframe agentic trajectories as paths through a shared embedding space of actions and environmental observations. We show that successful and failed rollouts of the same task separate in this space, with simple per-task discriminators predicting held-out rollout outcomes at up to 0.95 AUROC across three agent benchmarks and three policy models. To exploit this signal across tasks, we introduce Precog-RM, a transformer-based model that meta-learns a representation in which outcome prediction can be performed in a few-shot setting. We show that Precog-RM generalizes to unseen tasks and models, and gives confident predictions of trajectory success long before termination. Deployed as an early stopping controller for Group Relative Policy Optimization (GRPO) on Tau2-Bench, Precog-RM cuts training FLOPs by up to 47.3\% while maintaining performance comparable to baseline GRPO.

View on arXivView PDF

Code Implementations(1)

alexgill321/precog-rmCommunity71%
MIT

Precog-RM: outcome-predictive representations of LLM agent trajectories (Divergent Paths, arXiv 2026)

00Oct 12, 20261 weeks agoMIT

Similar Papers

TAPO: Transition-Aware Policy Optimization for LLM Agents

83% match2026

MacFormer: Map-Agent Coupled Transformer for Real-time and Robust Trajectory Prediction

83% match2023

VISTA: A Vision and Intent-Aware Social Attention Framework for Multi-Agent Trajectory Prediction

83% match2025

Co-Evolving Acting and Forecasting for Language Agents

83% match1 repo42026

Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal Representations

83% match2026

Multi-Objective Decision Transformers for Offline Reinforcement Learning

82% match2023

Cite this paper

@article{gill2026divergent,
  title  = {Divergent Paths: Learning an Outcome-Predictive Representation of LLM Agent Trajectories},
  author = {Alexander Gill and Guram Gogia and Francisco Guzmán},
  year   = {2026},
  eprint = {2610.19614},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2610.19614v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2610.19614.svg)](https://codeofpaper.com/paper/2610.19614)

Discussion