How heterogeneously will a central-bank statement be read? I build an ex ante measure of FOMC communication-induced disagreement by simulating 30 heterogeneous synthetic traders who read each post-meeting statement and revise their rate expectations; the cross-sectional dispersion of those revisions is the measure. Constructed only from information public at release—the current statement, the previous regular statement, and pre-release market anchors—it predicts the 30-minute statement-window Treasury reaction: held-out rank correlations are 0.56 at the two-year and 0.41 at the ten-year maturity, and the relationship holds on the full sample of statement-only meetings with no projections release (0.46 at two years). In full-sample horse races the signal survives controls for high-frequency policy surprises, Summary of Economic Projections releases, pre-announcement volatility, and the model’s own mean revision. The measure also tracks independent human disagreement: across 2000–2025 it co-moves with cross-forecaster dispersion in the Survey of Professional Forecasters (0.58 at the front end), and it forecasts movements in market-based policy uncertainty. Sentence-level decompositions attribute the disagreement to balance-sheet language rather than to the rate decision, an attribution that replicates within the measurement model but not across model generations. The evidence is reduced-form: a market-relevant, ex ante ranking of how contentiously a statement will be interpreted, not a causal estimate of communication effects.