Research

Paper

TESTING March 19, 2026

Reflection in the Dark: Exposing and Escaping the Black Box in Reflective Prompt Optimization

Authors

Shiyan Liu, Qifeng Xia, Qiyun Xia, Yisheng Liu, Xinyu Yu, Rui Qu

Abstract

Automatic prompt optimization (APO) has emerged as a powerful paradigm for improving LLM performance without manual prompt engineering. Reflective APO methods such as GEPA iteratively refine prompts by diagnosing failure cases, but the optimization process remains black-box and label-free, leading to uninterpretable trajectories and systematic failure. We identify and empirically demonstrate four limitations: on GSM8K with a defective seed, GEPA degrades accuracy from 23.81% to 13.50%. We propose VISTA, a multi-agent APO framework that decouples hypothesis generation from prompt rewriting, enabling semantically labeled hypotheses, parallel minibatch verification, and interpretable optimization trace. A two-layer explore-exploit mechanism combining random restart and epsilon-greedy sampling further escapes local optima. VISTA recovers accuracy to 87.57% on the same defective seed and consistently outperforms baselines across all conditions on GSM8K and AIME2025.

Metadata

arXiv ID: 2603.18388
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-03-19
Fetched: 2026-03-20 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.18388v1</id>\n    <title>Reflection in the Dark: Exposing and Escaping the Black Box in Reflective Prompt Optimization</title>\n    <updated>2026-03-19T01:14:36Z</updated>\n    <link href='https://arxiv.org/abs/2603.18388v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.18388v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Automatic prompt optimization (APO) has emerged as a powerful paradigm for improving LLM performance without manual prompt engineering. Reflective APO methods such as GEPA iteratively refine prompts by diagnosing failure cases, but the optimization process remains black-box and label-free, leading to uninterpretable trajectories and systematic failure. We identify and empirically demonstrate four limitations: on GSM8K with a defective seed, GEPA degrades accuracy from 23.81% to 13.50%. We propose VISTA, a multi-agent APO framework that decouples hypothesis generation from prompt rewriting, enabling semantically labeled hypotheses, parallel minibatch verification, and interpretable optimization trace. A two-layer explore-exploit mechanism combining random restart and epsilon-greedy sampling further escapes local optima. VISTA recovers accuracy to 87.57% on the same defective seed and consistently outperforms baselines across all conditions on GSM8K and AIME2025.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.MA'/>\n    <published>2026-03-19T01:14:36Z</published>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Shiyan Liu</name>\n    </author>\n    <author>\n      <name>Qifeng Xia</name>\n    </author>\n    <author>\n      <name>Qiyun Xia</name>\n    </author>\n    <author>\n      <name>Yisheng Liu</name>\n    </author>\n    <author>\n      <name>Xinyu Yu</name>\n    </author>\n    <author>\n      <name>Rui Qu</name>\n    </author>\n  </entry>"
}