Mechanistic Interpretability Explainer
Activation Patching matters when a deployment owner needs more than a pass/fail eval. It gives the audit team a way to collect internal evidence, test a causal hypothesis, and state the limits before changing a production control.
Field guide
Test whether a candidate internal component changes behavior when swapped or ablated.
Does changing the suspected feature, layer, head, or residual stream state change the model answer in the predicted direction?
Causal test memo showing effect size, negative controls, and cases where the hypothesis fails.
Patch results can be brittle, distributed mechanisms can be missed, and prompt-pair design can bias the finding.
Evidence priority
Must be collected before the interpretation is trusted.
Use as supporting evidence and keep unresolved ambiguity visible.
Use as supporting evidence and keep unresolved ambiguity visible.
Use as supporting evidence and keep unresolved ambiguity visible.
Audit template
