Many important economic outcomes result from the cumulative effects of smaller choices and events, so the best outcomes require, minimally, accounting for total outcomes so far. We formally show that narrow bracketing — the neglect of such accounting — is costly and identified if and only if the willingness to pay for an option varies across all single choice sets, unifying, extending, and generalizing prior results. We empirically document narrow bracketing in work choices in a pre-registered experiment on Amazon Mechanical Turk: bracketing due to separate or combined choice presentations changes average reservation wages by 13-28%. In our experiment, broad bracketing is so simple to implement that narrow bracketing is hard to reconcile with optimal conservation of cognitive resources. An attempt at debiasing by drawing attention to the interdependencies has limited success.