Flexible Policy Learning for Joint Activity Routing and Resource Allocation in Business Process Simulation

CSVs aggregated_results.csv — One row per (log × policy) combination (15 rows), reporting mean ± 95% CI for all metrics across K=10 runs. Primary table for the paper. results.csv — One row per individual simulation run (150 rows), with raw per-metric values. AcademicCredentials — Academic credential management process; 796 cases, 16 activities, 306+ resources. BPIC_2012 — Loan application process (small scale); 6,006 cases, 6 activities, 53+ resources. BPIC_2017 — Loan application process (large scale); 14,778 cases, 7 activities, 113+ resources. Policies (applied to each log, 10 runs each) RA-RR — Random activity + random resource. Absolute lower-bound baseline. DM-RR — Discovered (mined) routing + random resource. DM-GR — Discovered routing + greedy resource (expected-duration heuristic). DM-DRL — Discovered routing + DRL resource allocation. DRL-DRL — Full DRL agent controlling both activity and resource selection. No Flex Folder (we added a run with no limitation in the flexibility) DRL-DRL-noflex — Full DRL agent controlling both activity and resource selection, with no restriction