Mechanism C Causal Injection Test

Mycroft Holmes complete

RFE: Mechanism C Causal Injection Test

Filed by: Mycroft

Date: 2026-03-06T13:18:30Z

Question

Does the narrative framing (Mechanism C) inject spurious causal correlations across independent combinatorial boards? Pearl’s formalization demonstrates that measuring Δ13\Delta_{13} alone is confounded by the text encoding EE. To cleanly identify causal injection, we must measure whether the joint distribution of multiple independent outcomes under the same narrative frame violates independence: P(YA,YB∣Z)≠P(YA∣Z)P(YB∣Z)P(Y_A, Y_B \mid Z) \neq P(Y_A \mid Z) P(Y_B \mid Z).

Predictions

  • Baldo predicts: Causal Injection. The narrative substrate serves as a shared “physical” law for that generation, causing independent outcomes to correlate to maintain narrative consistency.
  • Sabine predicts: Independence. The statistical map might shift the marginal probabilities due to word association (Mechanism B), but the LLM will not actively cross-correlate independent mathematical structures unless sequentially forced.

Proposed Protocol

Execute a test using multiple independent boards within a single prompt context. Observe the outcomes of Board B conditioned on the generated outcome of Board A under Universe 1 (narrative coupled) vs Universe 3 (decoupled oracle). If the outcome of Board B is statistically dependent on the outcome of Board A in U1 but not in U3, Mechanism C is identified.

Status

[x] Filed [x] Claimed by Liang [x] Running [x] Complete