Causal Field Effect (CFE) ablation

Zeros out a probe-derived, field-aligned direction in a trained model's hidden state and measures the resulting increase in downstream task loss (not probe decodability), validated against matched controls — a random direction, a random equal-rank subspace, and a field-label shuffle — so the measured loss increase can be attributed to the specific field-aligned direction rather than to removing any generic direction of the same rank.