Dataset and analysis code for AI sycophancy, epistemic calibration, and verification in university students' evaluation of research proposals

De-identified quantitative dataset and reproducible Python code for a balanced 2×2 between-participants experiment with 120 first-year Aerospace Engineering students, crossing conversational tone (praising vs. neutral) with epistemic stance (validation vs. calibrated assessment). Each participant evaluated four aerospace research topics under a single condition and provided one global evaluation. The package reproduces the descriptive statistics, factorial ANOVAs with partial eta squared, Levene and Shapiro–Wilk checks, Benjamini–Hochberg FDR adjustment, Tukey HSD, cumulative-logit proportional-odds sensitivity models, and the categorical final-decision analyses (chi-square, Cramér's V, and 100,000-permutation p-values) reported in the associated article. Aggregated category frequencies from the qualitative coding of the open-ended responses are also included (counts only). The raw free-text responses are not shared to protect participant privacy. Contents: de-identified participant-level data (CSV and XLSX), analysis code, generated output tables, codebook, and reproducibility metadata.