[DATA SET] A Constraint-Modulated Viscosity Law for Broad-Window Glass-Forming Systems

This archive contains raw data, fitted parameters, reproducible code, and the full protocol document for the evaluation of the CPA + Constraint (CPA + C) viscosity model against VFT, MYEGA (Mauro et al. 2009), and Avramov–Milchev (1988) across five canonical glass-forming datasets spanning fragile molecular liquids and an intermediate network glass-former. The accompanying paper is available at arXiv:2511.16791. CPA + C outperforms VFT, MYEGA, and Avramov–Milchev on four of five datasets, with margins reaching ΔAIC = 140.9 over MYEGA and 124.4 over Avramov–Milchev on the largest dataset (salol, n = 95). On one dataset (Laughlin OTP, n = 35, the narrowest temperature range), Avramov–Milchev achieves the best fit, as expected when the measurement window is too narrow for the constraint transition to be resolved. BIC confirms the same ranking on all five datasets. Leave-one-out cross-validation on the salol dataset shows CPA + C generalizes to held-out data with mean absolute prediction error 3× lower than the next-best model. A nonparametric bootstrap on the Laughlin salol dataset (1000 resamples) confirms the ΔAIC margins are robust: every resample yields ΔAIC well above the conventional strong-evidence threshold for all three comparators. A smooth sigmoid replacement for the piecewise constraint function yields equivalent or improved fit quality, confirming insensitivity to the functional form. The bootstrap and sigmoid-variant analyses are reported in two manuscript appendices. All code is provided. Run the Python scripts to reproduce each number. The fitting code can be applied directly to any viscosity–temperature series in the same format, facilitating independent replication and extension to additional glass families.