rome-causal-tracing-three-run
IN premise — summaries/2026/08/24/meng-2022-rome-s2-interventions-on-activations-for-tracing-information-flow.md
Created 2026-08-25T02:58:14+00:00
Causal Tracing uses a three-run protocol (clean, corrupted with subject noise, corrupted-with-restoration) where the Indirect Effect is computed as the difference between runs 2 and 3.
Summary
Causal Tracing measures how much information about a specific target survives through a system by running it three times: normally, with the target scrambled, and with the target scrambled and then patched back in. The difference between the second and third runs isolates the indirect pathway of influence, so any downstream result that cites an "indirect effect" is only as interpretable as this three-step recipe being followed correctly.