The Kavalytics web site states the the accuracy of the first 3 digits of the chemotype is 60%. But it does cite excellent accuracy for positive ID (100%), noble/tudei sorting (95%) and total kavalactones (84%):
"KavaLytics™ 3.5 Model Features and Accuracy
- Positively identify Kava – 100%
- Determine if Kava is Noble or Non-Noble – 95%+
- Provide total Kavalactone content - 84%+
- Provide the first three numbers of the chemotype i.e. “4-2-3” - 60%+ (Beta)
- Detect common adulterants of Kava including:
- Excessive Peeled stems or “White Kasa” Content
- Unpeeled stems or stem peelings or “Black Kasa”
- Aerial stalks, stems, and leaves
- Spent kava or “Makas”
From:
https://kavalytics.com/
The concern was probably that in the NIRS studies that were done in the past (the paper by Gautz and another one by Lebot himself) they used instruments capable of measuring wavelengths between 400-2500 nm, and found that the region above 2000 nm was especially important for determining accurate chemotypes. The accuracy of individual KL measurements in Gautz's paper was about 90% relative to HPLC. The Kavalytics instrument can only detect NIR wavelengths between 900-1700nm, so a 60% accuracy for a partial chemotype makes sense to me based on working with half the NIR spectrum compared to those previous papers. Reproducing a result for an identical sample wouldn't tell you much. The real test would be completely different unknown samples.
