11.4 Falsifying a latent interpretation

Latent common causes can contribute to nonclosure by removing directions needed to express visible brackets. But the same signal can arise from target mismatch, ordinary domain shift, weak overlap, omitted observed coordinates, insufficient adjustment, or field error. Before a residual is called latent-sensitive, BRIDGE audits:

1. direct observational–regime distribution gaps;

2. overlap, density-ratio stability, and effective sample size;

3. held-out field continuity and calibration;

4. random, permuted, or null tangent controls;

5. sensitivity to adding measured adjustment coordinates; and

6. matched baselines for the actual data regime.