A fixed classifier. Results that remain comparable.

The classifier was calibrated on ten nationally representative Bulgarian surveys conducted between 2016 and 2026 (n ≈ 10,900 respondents). Cross-national validation (January 2026, online panels in Bulgaria, Greece and North Macedonia, n = 1,774) showed a structural correlation between profile signatures across the three countries of r > 0.99. A hold-out validation applied the 2026 classifier to a Bulgarian face-to-face sample from 2018 (n = 1,015) that was excluded from the calibration corpus. It shows a deviation of 2.2 to 5.3 percentage points from the empirical benchmark across an eight-year interval.

The technical principle of the frozen classifier is established in science. In microbiology it was applied by France et al. (2020) in VALENCIA, a nearest-centroid classifier for vaginal microbial communities published in Microbiome (BMC/Springer Nature). The authors describe exactly the same problem with k-means that we address: dependence on the specific sample and inability to compare directly across studies. In psychographic segmentation, PsychoProfiling™ applies the principle for the first time with a published mathematical specification and international academic validation.