Splits = sending duplicate to another lab for a double check. Results are expected to be with 30% of each other or you may have a data problem.
Level 3 data here has had sufficient Quality Assurance as to be true decision-making data.
Oh, we call it different way here (and I'd translate it differently) so it had not occurred to me.
Splits were used for previous round of testing - results were well within 10%, 5% most of the time. Decided to not use any external service this time.
I'm not a data scientist and I have no problem admitting I'm kinda lost there - I have no idea about English terminology, so please go slow on me. I'm no expert in all fields.
What things I done to ensure the data are correct - if I have missed something important, please, let me know, I'll be only happy to improve my services in the future:
6 point calibration curve of certified reference standard was established, almost perfectly linear,
bracket method of testing standard (establishing calibration curve before and after the unknowns were tested),
sample of known concentration tested along with unknowns [result was within 1% of real content(!)],
triplicates of each sample were tested [I think the RSD was on average not over 1%],
noise and drift were controlled according to ASTM standards,
standard and samples were tested immediately after preparation for quantitation,
orthogonal methods of quantitation were used
I might've forgotten to write down something, but any input is appreciated.