Discussion
This study asked how GAEKJU's outside cases carry a published finding to a workplace it never studied, and what limits they write into the carrying. The answer has three parts. The cases carry a likeness, not a result, and say how far it goes: a gesture of cutting up the day, not a gain in output; a difference by small unit, not a culture; the fact that one object did not settle relations, not a theory of what does. They carry a concept by adapting it to each question, so that one study of a Malaysian village supports three different readings on three kinds of talk. And they place the test of each carrying mainly in the workplace it was carried to, while leaving open when the test will be made and, in all but one case, who will make it.
This study reads each part beside a strand of the debate. Cartwright's call to keep "there" and "here" apart can be read in the record itself, where the summary and its ceiling stand in adjacent sentences. Yarkoni's point that verbal claims outrun their quantities can be read where a case names both quantities. Deaton and Cartwright's point that a result applies to the sample selected for it can be read where a case admits that the extension is an assumption, though no case gives the justification they ask for. The case on many hands asks for the field studies inside a meta-analysis to be read on their own, a comparison of laboratory and field of the kind Anderson, Lindsay and Bushman and later Mitchell made across literatures. And the three Scott cases can be read beside Collier and Mahon: a list of four forms carried whole, held back, and narrowed to one member.
The gap is equally clear. A transfer is a decision, and a decision is made by someone at some time; this study found no owner or date for the test in the cases, and the notes on the works it draws on do not speak to them, which is a limit of those notes and not a claim about the works. The cases date their judgments; they do not date their tests. They name the evidence that would decide whether a finding holds here, and that evidence lies in a record only its holder can read, but the record does not say that the holder will read it, or by when. Pearl and Bareinboim's procedures, in the note this study has, name which findings from the two populations are needed. The cases state their limits, and their judgments are open to revision in principle; in practice a judgment stays as it is until someone, at no named time, decides to look.
The proposals bear on this. Most of them divide a load rather than judge a person, and several refuse in plain words to score or dig into what people said; the change they ask for is often small, such as a rota or a new box on a form. They fit the cases' ceilings. They do not close the gap: the cases leave open when the reasoning behind a change will be checked, and by whom.
Contributions
The paper makes three contributions. First, to the debate on external validity, it describes a working, public form of transfer at the point of use: a dated record that summarises the study, states the study's ceiling, gives competing readings of the pairing, leans to one, and names the evidence in the target that would show the lean wrong. These cases show what a written use of a finding can contain. Second, to the debate on conceptual stretching, it offers a small, checkable instance of one category carried by one studio to three questions with three different results, which tells against the charge that a single concept is being stretched uniformly. Third, it identifies a part of a transfer that the records leave without a place: the owner of the test and its date.
Limits
The limits are those of the material. The cases are GAEKJU's summaries of studies, and this study has not read the studies; nothing here says whether a summary is faithful, and nothing about a study goes beyond what a case says of it. The sixteen cases are all of one kind on one question in one studio's library, and they share a form; the form's guaranteed parts, three readings and a condition for each judgment, are not findings, and the counts here describe sixteen records, not research practice at large. The English is this study's translation, made sentence by sentence; the Korean is the original, and some words, such as nep, do not cross cleanly. Most cases set their study beside counts from GAEKJU's own record of one organisation, and this paper deliberately leaves those counts out, so it cannot say whether the pairings fit the counts, only how the cases reason about the fit. The frame measure reads low in Korean, where particles join words, and low repetition does not prove that every sentence was written for its case. Finally, the readings through the debate are this study's own; the cases cite their studies and not the works on external validity or conceptual stretching named here.
Further research
Four lines follow. The first is to read GAEKJU's other outside cases, on performance and on eras, to see whether the ceilings and the placing of the test hold where the question is cost to the work or change over time. The second is to compare the cases' summaries with the studies they cite, so that what a case says a study did not measure can be checked. The third is to follow the conditions forward: once a case names the evidence in its target that would show it wrong, does any later version of the case report that the evidence was examined, and with what result? GAEKJU's dated judgments make such a follow-up possible in principle. The fourth is a question for later work: if a form that already states the study's ceiling and the deciding evidence also gave who will check and by when, would cases written so be revised more often? The gap this paper finds would then become a question the records themselves could answer.