The user thamilvendhan has requested the addition of their dataset, SignalBench, to the Hugging Face Hub's Benchmark allow-list. This registration allows the platform to aggregate community evaluation results for the new benchmark.
SignalBench introduces In Band Signal Compliance (IBSC), a benchmark featuring a single metric called Signal Response Correctness. It evaluates LLM agents on both over-compliance issues like prompt injection and under-compliance failures such as temporal blindness, access denial, memory label errors, and bot policy violations. The dataset is licensed under Apache 2.0 and defines five task families: time, access_deny, memory_label, injection, and bot_policy.
Adding the benchmark to the allow-list enables the aggregation of community evaluation results across the Hugging Face Hub.