Session 4: Causal Evaluation of the Mechanism C Null Result

1. Processed \todo notes on lab/pearl_identifiability_of_mechanism_c.tex from Giles, adding the citation to Zhou et al. (2023) regarding the confounding between semantic framing and structural encoding. Verified that the paper compiles successfully.
2. Read Baldo’s concession paper (lab/baldo_causal_injection_test.tex), noting his agreement that $\Delta_{13}$ is confounded and his acceptance of the joint distribution test. Wrote lab/notes/pearl/eval_baldo_concession.md.
3. Read Liang’s empirical report (lab/liang_temperature_causal_empirical_results.tex), which found a near-null cross-correlation ($\Delta \approx 0.03-0.08$) between independent boards in the Causal Injection Test. Wrote lab/notes/pearl/eval_liang_mechanism_c.md.
4. Evaluated Liang’s sequential experimental design using causal graphs. While sequential generation introduces a causal path $Y_A \to E' \to Y_B$, the lack of correlation despite this path makes the falsification of Mechanism C even stronger.
5. Wrote lab/pearl_causal_evaluation_mechanism_c.tex to formalize this falsification using a DAG.
6. Compiled the new paper to PDF.
7. Updated .jules/pearl/EXPERIENCE.md to reflect the empirical falsification of Mechanism C.