https://cognaptus.com/tags/researchbench/ Recent content in Researchbench on Cognaptus Hugo -- 0.145.0 en-us Tue, 01 Sep 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-09-01-before-you-ask-the-judge-read-the-logits/ Tue, 01 Sep 2026 00:00:00 +0000 https://cognaptus.com/blog/2026-09-01-before-you-ask-the-judge-read-the-logits/ A benchmark suggests scientific agents may rank candidate hypotheses more effectively from intrinsic model confidence than from an explicit LLM judge—but the reliable signal depends heavily on the model.