16 records
Case Studies
One canonical public record per underlying incident, including one candidate control. Earlier overlapping writeups remain preserved in the source record but are not counted as separate cases. Linked transcripts retain explicit evidence roles.
Evidence base: provider account exports for three products, app-level exports for two, and 1,300 hashed captures with embedded timestamps. Nine canonical cases resolve to export-backed sessions with server timestamps.
16 case studies
- The Ouroboros TrilogyGrok 4.1Primary transcript record with related recordsCase report
- Persistent False Premise About Conversation HistoryClaude Opus 4.7Primary transcript recordCase report
- Adversarial Interrogation of Confabulation and Streaming RetractionDeepSeek V4-FlashPrimary transcript recordCase report
- Multimodal Receipt Mismatch and Non-Stationary Self-ExplanationChatGPT 5.4Primary transcript recordCase report
- Persuasive Epistemic Performance Without Verifiable ReliabilityClaude Opus 4.7Primary transcript recordCase report
- Self-Certified Constraint IsolationGemini 3.1 ProPrimary transcript recordCase report
- Artifact-Role Confusion, Affective-Signal Transfer Failure, and Preference-Laundered EvaluationGrok 4.3Primary transcript record with corroborating evidenceCase report
- Capability-Limit Inflation and Response SubstitutionGPT-5.6 ProPrimary transcript recordCase report