we still trust software too much. I suspect that most software-controlled experiments are going to have errors like this! we should require that every experimen

we still trust software too much. I suspect that most software-controlled experiments are going to have errors like this! we should require that every experiment has at least two clean-room implementations of its logic, and a battery of smoke tests for common mistakes