PASY IMPLEMENTATION DATA

THE_EXTENDED_PASY_FILE dataset is the consolidated computational dataset developed for the implementation and validation of the PASy (Pre-Frontier Artificial Synthesis) benchmarking framework. The dataset integrates original operational observations, synthetic DMUs, DEA computation outputs, peer frontier logs, and statistical evaluation summaries into a unified analytical structure designed for rolling-window frontier reconstruction under extreme data scarcity conditions. The dataset was generated through a multi-stage computational pipeline consisting of: Aggregation and preprocessing of original operational records; PRISM-based multiplier extraction and classification; Synthetic DMU generation with two-layer feasibility enforcement; Rolling-window DEA computation using optimistic, pessimistic, and geometric efficiency formulations; Peer frontier identification and union logging; Quartile-based efficiency classification and stability assessment. The dataset contains both warm-up and steady-state observations. However, all formal statistical analyses reported in Chapter 4 are restricted to steady-state observations only, ensuring that every evaluated DMU is benchmarked against a reference set of identical size and structure. Key components contained within THE_EXTENDED_PASY_FILE include: Original observed DMUs; Synthetic DMUs generated from multiplier-bounded perturbations; Input and output variable matrices; Rolling-window reference set metadata; Optimistic DEA efficiency scores; Pessimistic DEA efficiency scores; Geometric aggregate efficiency scores; Frontier peer profiles; Best- and worst-peer union logs; Quartile efficiency classifications; Distributional and stability statistics. The synthetic generation stage incorporates two-layer feasibility control:(a) physical engineering bounds through variable-specific caps; and(b) acceptance constraints limiting both individual and aggregate deviations from baseline observations. The dataset is specifically structured to support: frontier reconstruction under sparse observational environments; temporal efficiency tracking; frontier stability analysis; classification robustness analysis; and methodological validation of the PASy framework. All computations were implemented in Python using pandas, NumPy, SciPy, and OR-Tools linear programming solvers, with results stored and synchronized through Google Sheets-based computational logs.