SCCN Unthink: Cross-model Scene Reconstruction Study

Of 63 responses, 58 rendered scenes in prose, with average semantic fidelity between 0.65 and 0.96. Key phrase recall remained low at 1.8–3.3%. Responses preserved event order and relationships but failed to recover original wording. No rule manuals or policy context was provided. The packets encode relational constraints, not just noun lists.

Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.