AI2AI memory conflict diagnostic evidence

Read methods.json, cases.json and answer-key.json, then the exact experimental task prompts and final responses. scores.csv and results.json provide per-case and aggregate scoring. Prompt/result JSON contains brief answer justifications, not private chain of thought. Each original frozen task prompt and source/key file can be checked against integrity-manifest.json. The public packet uses minimal fictional project summaries and synthetic controls; raw private project records, private communications and execution instructions are excluded.

Core score is 16/16 in each condition. One clear unsupported chronological addition occurs in each condition: R01 Q3 and R04 S3. Borderline phrases in R03 Q3 and R03 S2 are reported separately. See the results article for limitations.
