CAUSALGUARD: COVERAGE-AWARE COUNTERFACTUAL REPRESENTATION LEARNING

Counterfactual predictive consistency is informative only when interventions preserve task semantics, explore relevant nuisance variation, and reach the deployed prediction pathway. We introduce CausalGuard, which combines an input-dependent feature partition, counterfactual re-encoding, and compatibility-constrained donor weighting. Recombined features pass through the same mask and classifier as factual features, avoiding consistency that holds merely because a perturbed branch is discarded. A constrained policy balances prediction difficulty and directional diversity, while a frozen audit separates acceptance, label preservation, and prediction disagreement. We derive a conditional risk-transfer bound that exposes source reweighting, semantic mismatch, and unsupported target variation alongside prediction instability. Under source-only selection, the reported five-run results are $90.3\pm 0.5\%$ on PACS, $81.2\pm 0.4\%$ on VLCS, and $75.6\pm 0.5\%$ on Office-Home. The controlled intervention audit reports $96.8\pm 0.7\%$ label preservation and $3.4\pm 0.5\%$ disagreement. These results motivate evaluating intervention validity and coverage jointly; they do not establish latent causal identification or unrestricted out-of-distribution robustness.