ADP is an open specification for multi-agent consensus — calibration-weighted voting, reversibility-tiered thresholds, and structured belief-update rounds that make bad-faith participation legible over time.