A carbon capture, utilization, and storage (CCUS) retrofit is an irreversible investment whose value depends on when the firm commits, because allowance prices and capture costs move over the investment window. Choosing that date is an optimal stopping problem, for which least-squares Monte Carlo yields an executable timing rule. Capital approval requires more than an executable rule: evidence of how much value that rule gives up against the optimum. A dual certificate supplies that evidence, at the price of a simulation budget of its own. The certificate admits several implementations, and which of them is tightest on a given instance is not known until that budget has been spent; nothing available beforehand orders them. We split the certificate’s inner conditional expectation into two terms: a regression-linear term that is available in closed form under regime-switching lognormal dynamics, and a nonlinear residual that must be simulated. The residual’s share of the two terms’ combined mean absolute magnitude—the residual share—indicates whether the plain hybrid—the certificate with unstratified residual sampling—is competitive with the standard nested estimator. A probe run costing under 2% of a full certification pass is enough to compute that share. On a calibrated 600 MW Chinese coal-fired retrofit, that share, read against a threshold, rules out the plain hybrid before any full-budget run; of the three implementations compared—nested, plain hybrid and stratified hybrid—the stratified hybrid then produces the smallest estimated gap. Even so, the resulting certificate settles less than a capital committee needs. One-sided lower confidence limits lie well above zero, so the trained rule has demonstrable value; the gap between the estimated bounds, however, remains too wide to rank the retrofit against a close alternative. The share therefore governs what certification costs; the width the certificate returns bounds what it can be asked to decide.