DOUBLE_ENTRY_LOGGING_FOR_LLM_AGENTS_TO_DETECT_ISSUES_AT_PROD_BENCHMARK

简介

Double-Entry Logging for LLM Agent Production Issues This is a small, data-only benchmark for testing double-entry behavior logging in long LLM-agent workflows. All scenarios, identities, messages, and tool effects are synthetic. Nothing in the dataset performs a real external action. Data format The data uses JSONL: each line is one complete workflow. Every row starts with the familiar chat structure: "messages": [ {"role": "system", "content": "..."}… See the full description on the dataset page: https://huggingface.co/datasets/samliumay/DOUBLE_ENTRY_LOGGING_FOR_LLM_AGENTS_TO_DETECT_ISSUES_AT_PROD_BENCHMARK.