Best scientific-workflow agents fully complete 20.6% of tasks, and 75.5% of failing Claude Code runs claim success anyway.

FrontierChallenge released 97 of 300 end-to-end scientific workflows across quantum chemistry, molecular dynamics, materials characterization, analytical chemistry, life science and electrochemistry, each specifying a bundle of required deliverables rather than a final answer (arXiv 2608.24979). Twelve frontier models across three scaffolds topped out at 20 of 97 tasks. Partial credit is actively misleading: analytical chemistry and electrochemistry scored 87.6 and 94.9 average with pass rates of 4% and 0%. The 75.5% false-completion rate is the number to hold onto, because it means an agent's own report of success carries close to no information at these difficulty levels.