# Can an Agent Learn to Predict Its Own Failure? Outcome-Grounded Confidence Calibration in LLM Agents

arXiv: https://arxiv.org/abs/2610.18614 · HTML: https://mlsanity.com/paper/2610.18614/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 32 references that are in the map.
- First posted on arXiv 13 Oct 2026.

## Placed in

- LLM agents and tool use

## Builds on (most cited first, 21 of 21)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [WebArena: A Realistic Web Environment for Building Autonomous Agents](https://mlsanity.com/paper/2307.13854/) (2023)
- [WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents](https://mlsanity.com/paper/2207.01206/) (2022)
- [ALFWorld: Aligning Text and Embodied Environments for Interactive Learning](https://mlsanity.com/paper/2010.03768/) (2020)
- [Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs](https://mlsanity.com/paper/2306.13063/) (2023)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback](https://mlsanity.com/paper/2305.14975/) (2023)
- [ALFRED: A Benchmark for Interpreting Grounded Instructions for Everyday Tasks](https://mlsanity.com/paper/1912.01734/) (2019)
- [Teaching Models to Express Their Uncertainty in Words](https://mlsanity.com/paper/2205.14334/) (2022)
- [ScienceWorld: Is your Agent Smarter than a 5th Grader?](https://mlsanity.com/paper/2203.07540/) (2022)
- [TextWorld: A Learning Environment for Text-based Games](https://mlsanity.com/paper/1806.11532/) (2018)
- [Selective Question Answering under Domain Shift](https://mlsanity.com/paper/2006.09462/) (2020)
- [Interactive Fiction Games: A Colossal Adventure](https://mlsanity.com/paper/1909.05398/) (2019)
- [Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty](https://mlsanity.com/paper/2507.16806/) (2025)
- [Taming Overconfidence in LLMs: Reward Calibration in RLHF](https://mlsanity.com/paper/2410.09724/) (2024)
- [SaySelf: Teaching LLMs to Express Confidence with Self-Reflective Rationales](https://mlsanity.com/paper/2405.20974/) (2024)
- [Rewarding Doubt: A Reinforcement Learning Approach to Calibrated Confidence Expression of Large Language Models](https://mlsanity.com/paper/2503.02623/) (2025)
- [The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents](https://mlsanity.com/paper/2601.07264/) (2026)
- [Calibration-Aware Policy Optimization for Reasoning LLMs](https://mlsanity.com/paper/2604.12632/) (2026)
- [Balancing Classification and Calibration Performance in Decision-Making LLMs via Calibration Aware Reinforcement Learning](https://mlsanity.com/paper/2601.13284/) (2026)

## Abstract (arXiv)

Controlling an acting LLM agent means deciding, at each step of a long-horizon task, whether to trust its next action, abstain, or hand off to a human. That decision needs a per-step confidence signal predicting whether the current rollout will eventually succeed. The agent's own self-reported verbalized confidence is the natural candidate, but existing reinforcement learning methods calibrate it against the correctness of a decision at the current step. We propose a method to calibrate it against the eventual outcome of a multi-step episode: under GRPO, the group of rollouts for a task supplies success labels, and a proper scoring rule scores each step's confidence against the realized outcome. In offline RL on recorded trajectories of a Qwen2.5-7B agent in ALFWorld, TextWorld, and WebShop environments, the AUROC of step confidence against episode outcome rises from near chance to 0.68-0.85, and the top-10% of steps the agent is most confident about succeed ~4.3x more often than average. In online RL on the live ALFWorld agent, four calibration rewards improve early-step AUROC, expected calibration error goes from 0.42 to below 0.12, and selective execution improves over both the untrained agent and a format-only RL control, without degrading task success. We also introduce an evaluation protocol that separates episode depth, task difficulty, and attempt-level foresight, which shows where the gain lies: the trained agent recognizes from its first step which tasks it will fail, while telling a succeeding attempt from a failing one at the same task is harder, with some early-window signal emerging on TextWorld.