“Researchers and engineers consistently underestimate project timelines, often by a factor of 2x to 3x even in good cases.”