Checking the engines: quantitative error bar assessment for DFT-based property predictions