AutoResearchEval finds one failure every model shares.

100 real frontier research tasks across seven scientific domains, full lifecycle, 800 annotated trajectories, 45-pattern failure taxonomy (arXiv 2608.14905). The headline isn't a leaderboard, it's a shared deficit: agents can't check what they produced against what they found, revise when it doesn't hold, or question whether the path was sound. Held across all 8 harness-model combos including the strongest models. That points at a model-level limitation, not a scaffolding one, which means a better harness will not patch it.