Beyond prediction agreement: a decision-aware validation framework for quantitative spectroscopy using commercial FT-NIR textile analysis

Analytical validation of quantitative spectroscopic methods is commonly dominated by numerical agreement metrics, although practical decisions are often determined by predefined acceptance boundaries. This study applied a decision-aware validation approach to a fixed commercial Fourier-transform near-infrared (FT-NIR) instrument-software-calibration system using 200 randomly selected routine binary-fibre textile specimens. FT-NIR estimates were compared with results from standardized reference chemical procedures using complementary continuous-agreement metrics and specimen-level conformity classifications. Lin's concordance correlation coefficient was 0.9836, the mean FT-NIR-reference difference was -1.02 percentage points, and the 95% limits of agreement were - 5.26 to 3.22 percentage points. Decision agreement was 91.5% (183/200). All reference results were conforming; the 17 discordances therefore represented additional FT-NIR rejections, of which 16 involved crossing the lower conformity boundary. Thus, the observed aggregate agreement metrics did not fully describe specimen-level decision consistency, which also depended on the direction of the difference and its position relative to the applicable boundary. Exploratory offset correction reduced average error metrics but left the width of the agreement interval largely unchanged. Post hoc diagnostic measurements identified specimen composition and fabric heterogeneity as candidate sources of variability for future study. The findings show how continuous method-comparison evidence and boundary-dependent decision outcomes can be evaluated jointly during post-deployment validation of quantitative spectroscopic systems.