AI TrackerFrontier desk · AEST
Back to news
normalaitech

RATED: Reliability-Aware AI Agent Tool Retrieval via Execution Dynamics

arXiv cs.AI · 10 Oct 2026, 03:38 AEST

RATED is a new framework designed to improve the reliability of AI agent tool retrieval by combining semantic matching with dynamic reliability adaptation. It significantly enhances performance in real-world scenarios, reducing fault selection rates during API outages.

Key facts

Read aloud

RATED is a new framework designed to improve the reliability of A I agent tool retrieval by combining semantic matching with dynamic reliability adaptation. It significantly enhances performance in real-world scenarios, reducing fault selection rates during A P I outages. Existing methods optimize semantic relevance but largely ignore tool availability during execution. RATED employs a frozen dual-encoder with contrastive fine-tuning for semantic matching, and a lightweight contextual bandit module that re-ranks candidates based on real-time execution feedback. Experimental results demonstrate that RATED achieves a fifty two point two percent relative MRR improvement over the ToolLLM A P I retriever on real-world queries. RATED reduces fault selection rate from eighty six point two percent to two point nine percent under severe A P I outage scenarios with a thirty percent failure rate, while incurring negligible latency overhead of less than one ms over encode-only inference.

Sources in this cluster