An AHP-MAP Framework for Transparent and Evidence-Calibrated Evaluation of Technology-Transfer Performance in New R&D Institutions

Evaluating technology-transfer performance in new R&D institutions requires an evaluation system that remains auditable while allowing empirical calibration with observed outcomes. This study proposes an AHP-MAP framework that combines the analytic hierarchy process with maximum a posteriori weight calibration. The primary and local AHP weights define the modes of Dirichlet priors, while a two-part occurrence–magnitude model links composite scores to zero-inflated and long-tailed monetized technology-transfer outputs. A common prior-strength parameter is selected from a prespecified candidate set through stratified four-fold cross-validation. Using 23 indicators from 36 institutions, the proposed method improves post-selection out-of-fold performance relative to the fixed AHP baseline. The Spearman correlation increases from 0.655 to 0.712, the area under the receiver operating characteristic curve increases from 0.742 to 0.805, the root mean square error for positive-output magnitude decreases from 1.481 to 1.453, and the mean joint negative log-likelihood decreases from 2.006 to 1.974. Compared with the purely data-driven model under the same formulation with κ=0, AHP-MAP remains closer to the original AHP structure and avoids excessive weight concentration. Its additive structure also enables exact decomposition into dimension-level and indicator-level contributions. Multiple-initialization, transformation, bootstrap, jackknife, and Monte Carlo analyses provide complementary evidence on weight and ranking stability across alternative analytical settings and sample sizes. The framework provides a transparent and auditable basis for performance assessment, weight adjustment, and institutional diagnosis.