# Instructions Are Evidence, Not Objectives: Measuring Outcome Alignment of Agents

arXiv: https://arxiv.org/abs/2610.17757 · HTML: https://mlsanity.com/paper/2610.17757/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use.
- Builds on 23 of its 32 references that are in the map.
- First posted on arXiv 13 Oct 2026.

## Placed in

- LLM agents and tool use / [Software engineering coding agents](https://mlsanity.com/agents/software-engineering-coding-agents/)

## Builds on (most cited first, 23 of 23)

- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [SWEET-RL: Training Multi-Turn LLM Agents on Collaborative Reasoning Tasks](https://mlsanity.com/paper/2503.15478/) (2025)
- [UserBench: An Interactive Gym Environment for User-Centric Agents](https://mlsanity.com/paper/2507.22034/) (2025)
- [DeepSWE: Measuring Frontier Coding Agents on Original, Long-Horizon Engineering Tasks](https://mlsanity.com/paper/2607.07946/) (2026)
- [SWE-chat: Coding Agent Interactions From Real Users in the Wild](https://mlsanity.com/paper/2604.20779/) (2026)
- [Learning to Ask: When LLM Agents Meet Unclear Instruction](https://mlsanity.com/paper/2409.00557/) (2024)
- [Ambig-SWE: Interactive Agents to Overcome Underspecificity in Software Engineering](https://mlsanity.com/paper/2502.13069/) (2025)
- [Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops](https://mlsanity.com/paper/2607.07663/) (2026)
- [HiL-Bench (Human-in-Loop Benchmark): Do Agents Know When to Ask for Help?](https://mlsanity.com/paper/2604.09408/) (2026)
- [Ask or Assume? Uncertainty-Aware Clarification-Seeking in Coding Agents](https://mlsanity.com/paper/2603.26233/) (2026)
- [SWE-Together: Evaluating Coding Agents in Interactive User Sessions](https://mlsanity.com/paper/2606.29957/) (2026)
- [SWE-INTERACT: Reimagining SWE Benchmarks as User-Driven Long-Horizon Coding Sessions](https://mlsanity.com/paper/2606.30573/) (2026)
- [EvoCode-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions](https://mlsanity.com/paper/2605.24110/) (2026)
- [Dialogue SWE-Bench: A Benchmark for Dialogue-Driven Coding Agents](https://mlsanity.com/paper/2606.13995/) (2026)
- [DiscoverLLM: From Executing Intents to Discovering Them](https://mlsanity.com/paper/2602.03429/) (2026)
- [Ambig-DS: A Benchmark for Task-Framing Ambiguity in Data-Science Agents](https://mlsanity.com/paper/2605.09698/) (2026)
- [ClarifyCodeBench: Evaluating LLMs on Clarifying Ambiguous Requirements for Code Generation](https://mlsanity.com/paper/2607.00711/) (2026)
- [Ask Early, Ask Late, Ask Right: When Does Clarification Timing Matter for Long-Horizon Agents?](https://mlsanity.com/paper/2605.07937/) (2026)
- [Asuka-Bench: Benchmarking Code Agents on Underspecified User Intent and Multi-Round Refinement](https://mlsanity.com/paper/2606.05920/) (2026)
- [SWE-Touch: Benchmarking Coding Agents When Users Touch the Code](https://mlsanity.com/paper/2608.02499/) (2026)
- [VibeLifeBench: Can Your Life Agent Be Proactive and Persistent in a Living World?](https://mlsanity.com/paper/2608.10875/) (2026)
- [A Unified Issue Resolution Benchmark for Requirement Clarification, Planning, and Code Generation for Coding Agents](https://mlsanity.com/paper/2608.09072/) (2026)
- [Interactive Task Alignment as a POMDP](https://mlsanity.com/paper/2607.16412/) (2026)

## Abstract (arXiv)

Large language model agents can follow complex human instructions to solve real-world tasks. In practice, however, initial instructions are often incomplete. In our analysis of 100 user-agent trajectories from SWE-chat, 56% of initial instructions omit requirements that later emerge as necessary, and 52% of the useful follow-up requirements could have been inferred from the project. A user states requirements from their own understanding of the task, and these stated requirements often differ from the true requirements that completing the task imposes. An agent should therefore treat the instruction as evidence rather than as the objective, inferring the true requirements from the instruction and the environment and satisfying them even when the user is not yet aware of all of them. We call this ability outcome alignment. In this work, we construct TiGanEval, a benchmark of software engineering and research tasks. In every task the user genuinely lacks part of the requirements, each requirement stays discoverable from the code and runtime behaviour of the environment, and requirements the agent infers are separated from those the user adds after observing delivered work. These properties together make outcome alignment measurable. Evaluating eight frontier models including GPT-5.6-Sol, we find outcome alignment broadly missing: the models lose 53 to 67 points of success rate when moving from the full specification to the partial instruction, and consequence feedback across four submissions does not close the gap. Furthermore, a plan-mode harness cannot compensate for the missing ability of the model itself, which places outcome alignment among the abilities the next generation of models should be developed for.