A calculation run twice with the same answer has been shown to repeat, not to mean what its author meant. This paper asks what one set of decision records takes a small executed calculation to have checked, and what the records do where the calculation and their own arithmetic part. It reads thirteen authored cases from AEGIS, a studio of the RISCON family: every case whose run counts the work a change would save, net of rework, and whose situation or question speaks of rework. All thirteen are synthetic settings written in Korean, not real events; every figure, office and procedure in them is the author's assumption, and the English quoted is this study's own translation.
Three findings follow. One figure, the minutes of rework, is read in two ways: the run adds it to every case, while seven records subtract it once from the whole. In those seven the record states a saving and its own run reports an increase, and they are the only increases among the rule's 93 runs across AEGIS. Nothing in the seven sets the two results side by side: the run's interpretation and the recommendation carry the stated saving, and the replay reports a match. And the records ask next for measurement of rework, not a reading of the figure. Read beside the debates on verification and validation, reproducibility and forecasts of time and cost, the set checks that its arithmetic repeats and leaves open what its numbers stand for.