True detection limits in an experimental linearly heteroscedastic system. Part 1