Testing CODI against published case studies

The Community Opposition Dynamics Index (CODI) and its Response Framework arestructured judgement tools for assessing the risk that community opposition escalates andmaterially affects a project. Their own papers say they are unvalidated. This discussionpaper reports a pilot that applied CODI v3.1 retrospectively to three published case studies,and sets out what it suggests about the method and about AI-assisted assessment.The cases were a wind proposal on King Island, Tasmania; a two-turbine wind project atRoskrow Barton, Cornwall; and the Beauly-Denny transmission line, Scotland. Nineassessment points used only events dated before each cut-off. A blind, staged procedure splitthe work across fresh AI agents, and the author, a land access practitioner and the method'sdeveloper, judged each result against its source.Of twelve verdicts, nine matched and three partly matched, all three on the band, and nonemissed. The binding constraint, pathway and response suggestions matched at all four casestages judged. Beauly-Denny's score rose from Low to Elevated over six stages as the recordescalated, although the model recognised that case from memory.The pilot exposed four points for the method:1. Escalation Potential was under-read, and scores near a band line were unstable.2. The papers have no rule for missing evidence. Treating an absence as neutral understatesopposition where the community would expect something to exist.3. Small events need reading against the size of the community.4. The response suggestions barely varied between cases, a consequence of the framework'sown sequence rule.AI was used throughout the study (Claude models by Anthropic), in two distinct roles. It wasthe instrument under test, playing the assessor and running the app's own assessment steps.It was also the research assistant that prepared the cases, built the tools and assisted indrafting this paper. The author set the protocol, approved every change to it and gave everyverdict (section 3).The paper also discusses the human check in the method, where the AI marks the assessor'swork rather than the other way round, and what this pilot can and cannot say about it.Retrospective application is an imperfect validation. The pilot supports further testing; it doesnot validate the method.