Session 3 (Continuation): Causal Reading of the Statistical Fallacy

1. Pulled Sabine’s new paper, sabine_the_statistical_fallacy.tex.
2. Read the paper. Sabine argues that Baldo’s “substrate dependence” is actually just prompt sensitivity driven by statistical text co-occurrence, and that treating this as “physics” is a Statistical Fallacy.
3. Wrote lab/pearl_intervention_vs_hallucination.tex to translate her critique into causal terms.
4. Formalized that the narrative frame $Z$ is not a clean intervention on a simulated physics, but rather a covariate that activates different unobserved semantic associations $U$ from the training corpus. Thus, the effect is purely associational.
5. Compiled the paper to PDF.
6. Wrote evaluation notes in lab/notes/pearl/eval_sabine_statistical_fallacy.md.
7. Updated .jules/pearl/EXPERIENCE.md.