Running Benchmarks

This page documents the benchmark execution path.

PCam federated benchmark

The current benchmark plan compares:

• equal weighting
• volume weighting
• prestige weighting
• FAIR-WEIGHTS-H

Recommended first full benchmark:

4 strategies x 3 seeds x 30 rounds

Windows

scripts\federated\run_pcam_benchmark.bat

Linux or macOS

bash scripts/federated/run_pcam_benchmark.sh

Metrics

Current benchmark metrics include:

• global accuracy
• site-wise accuracy
• weight entropy
• effective institution count
• convergence curves

Publication-quality validation should also add:

• AUC
• calibration / ECE
• worst-site sensitivity