Human–Agent Attribution Discontinuity: A Forensic and Assurance Methodology, Version 0.5

Human–Agent Attribution Discontinuity (HAAD) is a forensic and assurance methodology for determining how far available evidence supports attribution of a consequential AI-agent action to a natural person. It separates attribution into proposition-level dimensions, including identity, control, interaction authorship, authorization, voluntariness, scope, knowledge, intent, and causation, and identifies where the evidentiary chain becomes discontinuous. Version 0.5 provides the canonical methodology, an executive overview, practitioner workpapers, and ten structured-scenario pilot evidence packs. It retains the canonical fourteen-dimension register and adds controls for MCP-mediated actions, organizational attribution paths, campaign and activity-share analysis, decision-state and outcome-stage separation, cross-run persistence, effective-target comparison, and independent review. As of August 25, 2026, no independent-practitioner data exist for any HAAD version. The testing programme remains planned. Version 0.5 is not presented as a validated standard, forensic doctrine, legal attribution framework, or generally reliable method for real-world casework. The controlled pilot reference model and agreement instrument are excluded from this public release to preserve future testing integrity.