A grade quoted in the wrong unit is the cheapest way to be wrong by a thousand. One gram per tonne is one part per million — the same number, and worth knowing cold.
A certified reference material has a certified value and a standard deviation that came from many laboratories. Insert it blind in the sample stream and see whether the results come back where they should. Two standard deviations is the usual warning, three is a failure — and a run of results all on one side is worse than a single wild one, because it is a bias rather than noise.
Two analyses of what should be the same material. The usual measure is the half absolute relative difference — the gap between them over their mean, halved — and the usual rule of thumb is that ninety per cent of field duplicates should fall under about twenty per cent, coarse rejects under ten, and pulps under five.
Barren material put through the same crusher and the same laboratory. It should come back at nothing. The usual failure is a blank that follows a high-grade sample, which is not a laboratory error at all — it is the crusher, and it means the high sample before it may have carried grade into the one after.
A result reported as "<0.01" is not zero and it is not 0.01. The convention that does least harm is half the detection limit, and the honest thing is to say how much of the data that convention is inventing.
Every resource that was later written off had assays behind it, and the assays usually had no independent check inserted at the time. Standards, blanks and duplicates cost a few per cent of an assay budget and they are the only evidence that the numbers mean anything. Insert them blind, at a steady rate, and read them as they come back — not at the end.