Identification

Standard election forensics cannot separate genuine polarization from manipulation. This paper uses the polling stations a losing candidate formally challenged as a treatment group, comparing their forensic properties to the unchallenged remainder of the same election — turning the fraud allegations themselves into the counterfactual.

Across 85 quarters (2004–2025) and seven identification strategies, only Cholesky recursive identification survives in an administered-rate economy; the peak GDP response to a 100 bp hike is −0.195 pp. An LLM-classified central-bank tone instrument passes relevance but fails exogeneity.