Corrected version. E001 tested declarative vs narrative content structure on Perplexity, January 2026. Corrected result from reconstructed counts: 18/30 (60.0%) declarative vs 11/30 (36.7%) narrative, a 23.3pp gap, two-proportion p = 0.07, not significant at 0.05, clearing the lab's 22.0pp measured noise floor (E016) by 1.3 points. Exploratory 15-query baselines on ChatGPT (8/15 vs 5/15) and Gemini (9/15 vs 5/15) are directionally consistent and individually non-significant, reported here for the first time. Version 1's figures (n=75, 61% vs 37%, 24pp, p < 0.01) were transcribed from a write-up later found to contradict its own revision history and are withdrawn, along with v1's blinding, session structure, and session-variance claims, which have no surviving source. No raw data files exist. Full provenance in the paper.
Artur Ferreira (Fri,) studied this question.