A net saving, in the twelve authored AEGIS records read here, decides which way the recommendation points and decides nothing else. The records know this. They write down, beside the figure, the quality they cannot count, the costs they have only assumed, and a middle course whose worth lies in what it would teach. The paper's contribution is to show, sentence by sentence and on synthetic records that say what they are, where a designed record draws the edge of its arithmetic and how its words and figures part at that edge.
Discussion
The four debates read in the introduction each find something in the set, and none finds all of it.
The normative rule that only costs still to come should count, which Arkes and Ayton, The Sunk Cost and Concorde Effects (1999), set against the sunk cost effect, fits the records' subtraction, in this study's reading: nothing already spent enters the line. But such a rule says which costs to admit, not how to treat what cannot yet be costed, and that is where the records do their most visible work: they name such matters as unknown and keep them outside the line.
Commensuration, as Espeland and Stevens, Commensuration as a Social Process (1998), define it, is the comparison of different entities by a common metric. The records perform it on parts, buildings, maintenance, purchases, courses, care, supplies and sites. What the set adds to that picture is the record's own second column. In this study's reading, the records do not claim that the common metric holds everything. They put the incommensurable beside it, in the situation, in what is unknown and in the reasons for the middle option, and let the metric carry only the direction.
The make-or-buy literature asks the choice of arrangement to turn on more than price. Williamson, The Economics of Organization (1981), and Brown, Transaction Costs and Institutional Explanations for Government Service Production Decisions (2003), put the attributes of the transaction and the costs of managing it at the centre; Hart, Shleifer and Vishny, The Proper Scope of Government (1997), add that where contracts are incomplete, cost and the quality that cannot be contracted pull against each other; Domberger, Contracting Out by the Public Sector (1997), names contractual incompleteness and the ownership of physical assets among the conditions that may impede efficient contracting. The records ask, as their next inputs, for prices: statements, quotations, estimates. Only 4 of the 12 ask for anything on quality, reliability, performance or protection. The attributes these works place at the centre appear in the records, but as unknowns and as doubts, not as inputs to the line. This is a description of the authored set and not a claim about how any public body chooses.
The literature on contracting out and its reverse, read through Hefetz, Privatization and Its Reverse (2004), finds movement in both directions and gives monitoring a central place. The set has both directions too: three records ask about turning work over to a contractor, three about bringing it in or making at home, and the same subtraction serves both. The records' promises to watch afterwards are promises to compare costs with the calculation; quality returns only in a few lines that say when the question would be opened again.
The value of waiting, which Dixit, Investment and Hysteresis (1992), draws from the behaviour of firms under uncertainty, is not a claim this paper makes about the records. It supplies a name for the one thing every record offers and no run prices: the middle course, partial, staged or piloted, whose reasons speak of a smaller risk and a later or smaller saving. In two records, the condition for opening the question again is written for that unpriced course, while the mark recommends the full switch.
Contributions
The paper makes three contributions. First, it shows on one record set that an executed net-saving calculation settles the direction of every recommendation, the negative case included, and settles nothing about the comparison among the three options each record offers. Second, it locates what stands outside the line: quality, further costs and the terms of contracts named as unknown, and a middle option argued in words and never given a figure. Third, it identifies three places where the records' words part from their figures: units of time attached to some terms and not others, one label that names a part in the situation and a total in the run, and thresholds for reopening that sit at the point where the saving vanishes, at the present total, or at a figure whose reading the record leaves open.
Limits
The cases are twelve, all from one programme of one studio, and all authored synthetic settings whose costs, units and schedules are the author's assumptions. The paper describes how the records frame a choice; it does not say what any option is worth, which reading of a figure is meant, or what anyone should decide. The English is this study's own translation of Korean records, and a choice of word, such as rendering 억원 as "hundred million won" or keeping an ambiguity in the first case's question, could shade a reading; the Korean original is named in each case's seed. The counts rest on rules over Korean words, stated beside each count, and a different rule could sort a few records differently. The frame measure under-counts in Korean, where particles join words. The works are read from their published abstracts, not in full, and the paper says of them only what those abstracts support; where it goes further, it says so.
Further research
Three lines follow from the findings. The 101 other runs of the same rule stand in 22 other programmes of AEGIS; reading them would show whether the pattern of a signed figure, an unpriced middle course and unknowns kept outside the line holds beyond this programme, or belongs to its form of record. The records' own reopening thresholds could be read against their runs across all of AEGIS's calculation rules, to see where else a threshold sits at the present total rather than at the point where a result changes sign. And the middle options could be studied as a set in their own right: what each would need to learn, and what figure, if any, a record of this kind could give to learning before committing. Each would extend the contribution stated at the start: a reading of where designed decision records draw the edge of their arithmetic, and of what they write down beyond it.