So, starting with an analogy. Suppose you want to know the distance from Grand Central Station in New York City to a cafe a block from the Eiffel Tower. You h

So, starting with an analogy. Suppose you want to know the distance from Grand Central Station in New York City to a cafe a block from the Eiffel Tower. You have two options of how to make the measurement. Option 1 is to directly measure the distance. Option 2 is to use a known distance between Grand Central Station and the Eiffel Tower, then add the distance from the Eiffel Tower to the cafe. Option 1 has fewer dependencies, but Option 2 is much easier and (assuming your reference measurement is correct) will give a much more accurate measurement. Similarly, any time you're making a high-precision measurement, you want to make it relative to something that's already known to a high precision. It doesn't need to be the same high-precision measurement each time, so long as it's something that gets you into the same ballpark as the thing you're measuring. That way, you don't need to make an absolute measurement, but only need to make a relative measurement. For the mud samples, we didn't actually care that it was the example same elemental composition in the handful of NIST samples we had, only that it had a known amount of the trace elements we were measuring. That way, instead of needing to explicitly measure detector efficiency, everything became a very simple scaling question of "(X counts in sample) / (Y counts in reference) * (Z known trace element in reference)".