The public benchmark generates deterministic copied-majority worlds and scores aggregation methods against hidden binary truth.
python -m benchmark --worlds 500 --seed 7SPECIFICATION.mddefines the task, generation regime, metrics, controls, and interpretation boundary.world.pygenerates the synthetic worlds.evaluate.pycomputes the metrics.decision-relative-independence-v0.1.jsoncontains constructed falsification fixtures, not a benchmark result.
Synthetic performance does not establish real-world lineage recovery or truth discovery. See the research map for protocols and preserved outcomes.