This paper asked what fifteen authored AEGIS cases say may be carried from a public record of the past into a present decision, on what conditions, and what they rule out. The answer has three parts. The cases separate a record from what is drawn from it and let it pass only on stated conditions. They rule out the neglect of a record more often than its literal copy, and they soften the copy where they allow it. And the original on which every condition depends is absent from the record set, which says of itself that it refers to no public record. All fifteen are synthetic training cases, not real events, and the records they describe are the author's assumptions; what follows concerns the design of a record set, not any history or document.
Discussion
The debate on learning from the past is usually conducted about decision makers: whether leaders reason well from analogies, whether organisations encode the right inferences, whether a policy that worked elsewhere will work here. A designed record set is a different object. It does not reason from a particular history; it sets out, in advance and for any history, which uses of a record are recommended, which are allowed and which are ruled out. The fifteen cases are such a design, and they can be read as a position in the debate.
On conditions, the position is the cautious one. Every recommendation carries the record over only after a comparison, within a scope, or in stages, and the situations name what in the record would limit its reach: the assumptions of an old clause, the age of a classic, the fixed inputs of a wargame, the reliance of a memoir on recollection. In this study's reading this is close to the support factors of Cartwright and Hardie, Evidence-Based Policy: A Practical Guide to Doing It Better (2012), and to the question Cartwright, Presidential Address: Will This Policy Work for You? Predicting Effectiveness Better: How Philosophy Helps (2012), puts at the centre of the address: whether evidence of efficacy there warrants a prediction of effectiveness here. The cases do not use those words, but their needed inputs pair the record with a survey of present conditions, as if the second were what makes the first usable.
On the direction of risk, the position is the confident one. By the keyword rule of section 1.1, the cases exclude neglect in 11 of 15 places and literal transfer in four, and where they allow the literal transfer as an alternative they write its safeguards into its reasons. Khong, Analogies at War (1992), and Levy, Learning and Foreign Policy: Sweeping a Conceptual Minefield (1994), would locate the larger danger in the readiness to apply the past. The cases locate it in the refusal to apply it. In this study's reading, which goes beyond what the works say, they sit nearer to the point in Rose, What is Lesson-Drawing? (1991), that searching for fresh knowledge is not normal and is stimulated by dissatisfaction with the status quo, and to the view of Levitt and March, Organizational Learning (1988), that organisations learn by encoding inferences from history into routines. This study reads the cases as routines for turning records into inferences on conditions.
The two positions fit together in one design. A set that fears neglect more than copying, and that answers copying by conditions rather than by exclusion, is a set that wants records used and used carefully. Dolowitz and Marsh, Learning from Abroad: The Role of Policy Transfer in Contemporary Policy-Making (2000), present a continuum for distinguishing types of transfer and take up its relationship to policy failure. This study's own reading, which the authors do not make, is that the fifteen cases place their recommended option between the copy and the refusal.
The third finding qualifies the first two. A discipline of conditions is only as good as the comparison it requires, and the comparison requires the original. The cases ask for it first, list the questions about its making that only it could answer, and make reopening depend in four cases on something found in it. But the record set does not hold it, gives it no date or place of release, and states that it refers to none. Michael Howard, The Use and Abuse of Military History (1993), a reprint of a lecture of 1961, argues that the benefits of finding out what really happened outweigh the doubts, and uses the criteria of width, depth and context. In this study's reading, the fifteen cases supply the context and leave out the finding out. March, Sproull and Tamuz, Learning from Samples of One or Fewer (1991), separate agreement on an interpretation of history from the correctness of that interpretation. In this study's reading, a template of this kind can secure agreement on the procedure; correctness needs the history, and the template does not contain it.
None of this is a defect in a training set that says what it is. AEGIS states plainly that the records are authored, that the judgment marks are not factual bases, and that the questions have not been compared with public records. The finding is about what such a set can teach. It can teach where to put a record in a decision and which uses to distrust. It cannot teach what a particular record says, and its one internal disagreement, whether a figure from a record is to be recorded against the original or as an author's assumption, shows the seam where the rehearsal of source criticism meets the absence of a source.
Contributions
This paper makes three contributions to the debate on lesson-drawing and the use of history in decisions. First, it shows how a designed record set encodes a position in that debate through the places of its options: what it recommends, allows and rules out. The fifteen cases recommend a transfer on conditions in every case, which in this study's reading aligns them with the literature on support factors and context. Second, it shows that the same set, by the keyword rule, places the larger risk in neglecting a record rather than in copying it, ruling out neglect in 11 cases and literal transfer in four, and repairing the allowed copy in its own reasons. In this study's reading, that places the set nearer to the accounts of lesson-drawing and organisational learning than to the sceptical accounts of analogy and inference. Third, it identifies a limit that a template of this kind cannot pass: it can stage the comparison with an original, list the questions only the original could answer, and tie reopening to what is found in it, but it cannot make the comparison, because the original is not part of it.
Limits
The study reads fifteen cases from one programme of one studio. They are synthetic, written by one author for training, in Korean, and their public records are assumed rather than cited, so nothing here describes how any real organisation uses any real record. The classification of options into holding back and carrying over is a rule on Korean words; it fits the fifteen as read closely, but one alternative falls under neither heading, and a different list of words would move individual cases. The English is this study's own translation, made once and labelled, and a reader of the Korean may weigh some words differently. The readings through the works rest on their published abstracts and summaries, and a reading that goes beyond those is marked as this study's. The fixed order of the author's marks, recommended first and excluded last, is the template's and was not read as a finding; nor were the three options and the condition for reopening that every case carries.
Further research
Three lines follow. The first is to read the other cases across AEGIS's library whose situations name a public or declassified record, of which there are many more outside this programme, and to ask whether the same ordering of neglect and copying holds where the record is not the programme's subject. The second is to place one of these templates beside a real, citable original and see what the recommended comparison yields when it can be made. The third is to compare the excluded options of other studios in the RISCON family where a case begins from a source, to see whether the preference for use over neglect is AEGIS's or the template's. In each, the contribution this paper offers, a way of reading a record set's position on learning from the past through what it recommends, allows and rules out, can be tested on records that are not synthetic.