1\section{Conclusion}\label{sec:conclusion}2The central question was whether counterfactual supervision of \emph{where} to edit a small latent workspace3improves the quality--compute frontier beyond uniform workspace computation and adaptive global recurrence.4On the evidence collected here the answer is no. The sparse editor is learnable and helps a frozen backbone5by a few percent of loss and two points of exact match when it is applied once or twice at every position, but6a global-recurrence adapter trained with the identical recipe helps at least as much at every budget; where to7write barely matters for fixed schedules; an exhaustive oracle shows large opportunity (up to8\routingoraclebeight{}\,\% of loss) that a value controller trained on 100k executed counterfactual labels9cannot capture, because the opportunity sits in high-uncertainty states where a wrong action costs more than a10right one gains and the controller's within-state ranking is close to chance; and at small scale the11per-action dispatch overhead erases the arithmetic savings of sparsity. The exact bypass holds bitwise, the12differential readout needs a scale-aware bootstrap, and the parameter and operation accounting of the proposal13is reproduced. The scientifically justified next experiment is not a larger model but a better label: a14controller trained to rank actions within a state (pairwise or listwise, with the state's mean gain removed)15on labels collected at the scale where the editor's arithmetic is not dwarfed by dispatch, with the16global-recurrence adapter as the baseline to beat.17