One Reflection Is Not Enough: Self-Correcting Autonomous Research via Multi-Hypothesis Failure Attribution

2026-06-30

AI Summary

This work proposes SAGE, a novel system that addresses the limitations of existing autonomous scientific agents, which often rely on unstructured self-reflection after experimental failures and consequently fall into local trial-and-error loops or lose contextual coherence. SAGE models failure recovery as structured causal diagnosis, employing a Multi-Hypothesis Failure Attribution (MHFA) mechanism to generate multiple evidence-backed explanations based on dynamic trajectory features. It further integrates root-cause severity assessment with deterministic intervention routing to enable hierarchical, precise repair. Explicit outcome grounding constraints ensure that reflections are traceable and interventions reliable. Evaluated across 12 topics spanning five scientific domains, SAGE increases the metric output rate from 42% to 92%, improves result quality from 5.00 to 6.75 (on a 10-point scale), and achieves significantly higher blind-review scores than AI-Scientist-v2 (52.0 vs. 48.2), substantially enhancing code development and execution reliability.

Abstract

Autonomous research agents can now draft hypotheses, write code, run experiments, and produce papers, but they remain brittle when experiments fail. Under the prevailing paradigm, failure recovery is usually delegated to a single free-form reflection: a rich trajectory of metrics, logs, and design choices is compressed into one verbal critique, which often leads either to localized trial-and-error or to hard pivots that discard useful context. We propose SAGE, a Self-correcting, Autonomous, Grounded Experimenter, to tackle this failure-recovery bottleneck. Its core mechanism, Multi-Hypothesis Failure Attribution (MHFA), treats recovery as a structured causal diagnosis. By analyzing dynamic trajectory features, MHFA systematically generates multiple evidence-grounded explanations for a failure, independently evaluates their severity, and deterministically routes the verified root cause to the correct intervention level (hypothesis, experimental design, or implementation). To guarantee scientific honesty, SAGE further employs a grounded reporting mechanism that explicitly constrains drafted results to actual measured values, redacting hallucinated numbers. On a 12-topic, 5-domain benchmark, SAGE increases metrics-bearing outputs from 42% to 92% over a reflection baseline, improves artifact quality from 5.00 to 6.75/10, and blindly outscores AI-Scientist-v2 (52.0 vs. 48.2), with gains concentrated in code development and execution. While fully autonomous scientific writing and generating conference-ready papers remain notoriously difficult open problems for the entire field, SAGE successfully produces significantly more reliable and higher-quality scientific artifacts. Ultimately, by coupling structured recovery with explicit grounding constraints, SAGE significantly outperforms monolithic reflection paradigms, establishing a highly trustworthy foundation for future autonomous research.

Problem

• autonomous research agents
• failure recovery
• reflection bottleneck
• scientific reliability
• experiment failure

Innovation

• Multi-Hypothesis Failure Attribution
• Autonomous Research Agent
• Structured Failure Recovery
• Grounded Scientific Reporting
• Causal Diagnosis