https://calibratedagents.com/blog/Research, benchmarks and essays on AI hallucination detection and AI answer verification.enhttps://calibratedagents.com/blog/microsoft-decision-1-benchmarkhttps://calibratedagents.com/blog/microsoft-decision-1-benchmarkSat, 10 Oct 2026 00:00:00 +0000Microsoft launched Microsoft-Decision-1 on 9 October 2026. The next day, we put it through the same test as the other leading decision APIs: one judge call per answer, on the same LLM-AggreFact answers, measured for speed, accuracy and cost. No single provider wins everywhere.https://calibratedagents.com/blog/hallucination-detectionhttps://calibratedagents.com/blog/hallucination-detectionFri, 09 Oct 2026 00:00:00 +0000An AI answer is rarely all wrong. Usually one or two sentences are, and an answer-level hallucination detection check can't tell you which. We measured that gap on 5,916 human-labelled sentences from two public benchmarks, and how far checking each sentence closes it.https://calibratedagents.com/blog/hallucination-detection-benchmarkhttps://calibratedagents.com/blog/hallucination-detection-benchmarkThu, 08 Oct 2026 00:00:00 +0000Most hallucination detection benchmarks compare detectors with whatever thresholds their authors chose. Teams don't work that way: they label a handful of their own answers and tune on those. So that is the test we ran, on RAGTruth++ and LLM-AggreFact, with the same labels for every system.https://calibratedagents.com/blog/why-not-an-llm-judgehttps://calibratedagents.com/blog/why-not-an-llm-judgeTue, 08 Sep 2026 00:00:00 +0000LLM-as-a-judge has good recall and no precision control. In production, where most answers are correct, that turns into a flood of false positives. Here is the arithmetic, and what a calibrated verifier returns instead.