Who’s to Blame? Causal Responsibility Attribution for Macro Placement

When macro placement causes routing design-rule-check (DRC) failures, engineers must decide which macro to move. Interpretability is often proposed as the way to extract such actionable knowledge from a learned congestion model; we test whether what the model highlights corresponds to knowledge about the design at all, using a fully interventional ground truth. We evaluate thirteen attribution methods in five groups: causal (Halpern-Pearl responsibility and blame, probability of necessity, Shapley value, counterfactual effect), economic (Pigouvian externality price, LP-dual shadow price), gradient saliency on a learned congestion surrogate (Grad-CAM, Integrated Gradients, Input$\times$Gradient), a proximity heuristic, and a random control. We evaluate all thirteen against an interventional oracle, a small but real analytical-placement plus negotiated-congestion-routing flow in which every $do(\mathrm{macro}\ m \to \mathrm{site}\ s)$ query is a full re-place and re-route ($\sim$150 ms). Across three synthetic blocks (1,260 interventions, 2,991 place-and-route runs) we measure diagnosis (rank agreement with the tool's own but-for effects), repair (real DRC reduction after three guided ECO moves, on two disjoint candidate-site groups), and faithfulness (deletion curves). Economic and causal methods agree with the tool (mean Spearman $\rho$ $+0.67$ and $+0.29$); saliency is anti-correlated ($-0.05$, below random). Acting on causal rankings recovers 85-92% of the oracle repair bound and gives identical gains on both held-out site groups, while Grad-CAM collapses from 21.7% to 6.4% and the Pigou price from 23.8% to 15.9% when the available repairs change. The setting is a rare one for interpretability research: a domain where every counterfactual can be checked by re-running the real generative process, so claims of "discovered" knowledge are directly falsifiable. Saliency's failure here is a negative result with a constructive replacement: interventional responsibility over the feasible action set.