The thirteen synthetic inquiries of Dionyas give up almost everything a vignette is made of, the description of a situation, and keep two things: a role and a question. What they add in return is a sentence that says how the question could be tested, and in about half of them, what result would count against the question's own premise. That sentence answers, at the level of the question, the objection that an invented figure can be neither verified nor falsified. It does not say who would run the test, and its form tracks where the record came from more than what it is about. This chapter shows each step and states, at its end, what it adds to the debates on vignettes and personas.
1.1Data and method
The data are Dionyas's public list of inquiries, fetched by plain requests and kept as snapshots in the study's folder. The list holds 1,539 inquiries across 32 domains and gives its count of generated judgments as 0. Each inquiry sits in one of three lanes: 1,276 in the evidence lane, 250 in the qualitative lane and 13 in a symbolic lane. Each record has six parts this study reads: a title, a question, a scope, a list of needs, a boundary and a state.
The selection rule is one sentence: every inquiry in the qualitative lane whose scope calls it a synthetic question. The rule keeps 13. Across the whole list, 19 inquiries call themselves synthetic, and all 19 say in their boundary that their source card was a model's draft checked and registered by RISCON; no other inquiry says so. The rule leaves out the 6 synthetic inquiries of the evidence lane, and it left them out without reading them first. The thirteen spread over 11 of the studio's domains, with two each on AI and on live shows and one each on the others. None names a real artist, company or person.
The records are in Korean. Every quoted sentence below is from one English translation made for this study, sentence by sentence, with nothing added; the appendix gives all thirteen in full. Word counts and word tests run on the Korean.
The frame was measured before anything was read as evidence. Across the list, the scope, the needs and the boundary are mostly shared wording: measured by runs of five words that recur in at least one record in twenty, 46% of the scope, 30% of the needs and 41% of the boundary is frame. The thirteen are too few to reach that threshold on their own, so the measure was repeated inside the selection, counting a run as frame when it recurs in at least three of the thirteen. Inside the selection, 81% of the boundary and 50% of the scope is frame: every boundary has the same wording except for a card number, and every scope ends in the same tail of synthetic question, simulation and awaiting judgment. The question and the needs carry no frame at all, 0% each. What is a record's own, then, is its title, its question, the role that opens its scope, and its one need, which stands in no other inquiry of the list. Only these are read as evidence below. In the sections that follow, the opening of the scope is called the role and the one need is called the test line.
Three counts were made by word tests on the Korean, and they should be read as such: whether a test line carries a condition, whether it names a result that would weaken the premise of its question (words such as weakens, overrated, too much power, another layer, only trust was missing, the logic of technique), and whether it prescribes a procedure of listening, asking, writing down or setting side by side. The boundaries name each record's source card by a number with a short series prefix; the thirteen fall into three series, which this study calls A, B and C in the order the list first gives them, with 2, 5 and 6 records. Every number in the paper comes from a script that reads only the snapshots.
1.2A scene compressed to a role
A vignette, in the sense Cheryl S. Alexander and Henry Jay Becker, The Use of Vignettes in Survey Research (1978), gave the word, is a systematically elaborated description of a concrete situation. The Dionyas records do not elaborate. Their role is on average 4.1 words long in the Korean and their question 9.5. There is no paragraph of setting, no named character, no sequence of events. What stands in place of the description is a job:
The cover director of a fictional album team · synthetic question · simulation · awaiting judgment
The tail of that line is shared by all thirteen; the first words are the record's own, and they are all there is of the scene. Thirteen records give thirteen different roles. Eight of them work for a label, an agency or an entertainment company; four work with a new artist or a new label; one is not a worker at all but a regular in the audience at live shows; two say outright that their team is fictional. None has a name. The choice is consistent with the boundary's claim that the scene was set for the question, and it avoids, deliberately or not, the trap Tara Matthews, Tejinder Judge and Steve Whittaker, How do designers and user experience professionals actually perceive and use personas? (2012), found in personas, attributes that mislead or distract. A role with no name, age or taste has few attributes to mislead with.
What the role leaves out, the question must carry. In most records it does so as a question about a situation the role would face. In three it does not ask at all. One record's question is a plain sentence:
The choreography is known, and the voice is still unfamiliar.
Another reads like the first line of a story:
The interlude cut from the release sheet came back on the stage.
The third is an assertion with the shape of a worry:
Vocal correction that gets by when only half right: where the director's ear begins to grow dull.
These three are the only scene-like moments in the selection, and they are unusual in the list. Outside the 19 synthetic inquiries, none of the 1,520 others has a question without a question mark; among the 19, four do. The synthetic records are thus the only place in Dionyas's list where a question is allowed to become a situation, a state of affairs stated flatly and left for the reader to turn into a question. This study reads them as vignettes reduced to their opening line: Rhidian Hughes, Considering the Vignette Technique and its Application to a Study of Drug Injecting and HIV Risk and Safer Behaviour (1998), calls vignettes stories, and these are stories of one sentence.
The questions that do ask often ask about a place or a person rather than a quantity. Three ask who or whose, three ask where or how far. The regular in the audience is given this:
Whether it is a person on the stage or only the sound that is a person: where should the audience learn this before buying a ticket?
And the new reviewer is given a figure of speech in place of a description:
A bundle in which demo tracks pulled out with AI pile up neatly: how far should a new reviewer open it?
The bundle is the whole scene. It says that demo tracks made with AI pile up, and that a newcomer must choose how deep to open the bundle, without saying how many, from where or under what rule. Aguinis and Bradley's realism, a carefully constructed and realistic scenario, is not what these records aim at. They aim at a recognisable predicament stated as briefly as it can be. Whether that brevity costs them the validity Alexander and Becker claimed for concrete situations is a question the records cannot answer, because nobody has yet answered them.
1.3The test line
If the scene is thin, the test line is where the records spend their words: 18.3 on average in the Korean, nearly twice the question. Every one of the thirteen has exactly one, and none is shared with any other inquiry of the list. That alone sets them apart. Elsewhere in Dionyas's list the needs lean on shared wording, 30% of their text being frame, and records with a single need are a minority, 420 of the 1,520 that are not synthetic. Here the one line is written for the question it follows.
All thirteen lines are phrased as conditions: if this is observed, then that follows. The clearest of them do more. They name the result that would undo the premise of their own question. The record about the erased chorus asks whose song remains when the performance team erases a chorus the fans have memorized. Its test line answers with a procedure and a way to be wrong:
In another arrangement of the show, put the chorus back and listen for where the singing-along arises. If the silence is the same, the fans' memory was overrated.
The question assumes that the fans' memory of the chorus matters, that erasing it takes something from them. The second sentence says what would show that assumption to be too large: if the room is as silent with the chorus as without it, the memory was not doing what the question supposed. The test is aimed not at an answer to the question but at the question's footing.
Seven of the thirteen lines have this structure. The record on the cover asks whether a song can go against the mood its cover promised first, and then:
Ask listeners who have not seen the cover about the song's feeling, and set the words of listeners who have seen it beside theirs. If the difference is faint, too much power was given to the packaging.
The record on the fans' nickname asks to whom a word remains when an advertisement takes it, and supplies the observation that would weaken the reading of loss:
Listen to how the places where fans use the nickname, and its meaning, changed before and after the advertisement. If the use carried on, the reading that the advertisement took the word away is weak.
The record on choreography first does the same for the idea that the order in which people meet a team shapes what they remember:
Listen to what people who first met a live scene of the same song and people who first saw the choreography remember. If there is no difference in memory, the blame on the order of arrival weakens.
And the record on the outside composer whose demo keeps losing its way sends the inquiry elsewhere if its first suspicion fails:
Hold a meeting that hears only the song before seeing the chart, and write down the scenes that remain. If the assessment is the same, the reason the demo lost its way must be sought in another layer.
These five share a grammar. A first sentence prescribes what to do, and it is almost always to listen, to ask or to set two groups' words side by side. A second sentence says what a null result would mean, and the null result always tells against the question's own suspicion. Here the vignette's method has come back into the record by another door. Atzmüller and Steiner describe the experimental vignette as a set of short descriptions whose important characteristics are varied systematically across respondents; the test lines vary one thing, the cover seen or not, the chorus present or not, the choreography met first or the live scene, and compare the words that follow. The variation that a vignette study puts into its scenes, these records put into the test they ask someone to run.
Two of the seven reach the same end with a different turn. The record on the unopened bundle does not say the premise would fail; it relocates the problem:
If the assessment is reversed when the same demo tracks are played to two people in a different order, there was a standard and only trust was missing.
And the record on the layer the tool ate names a case that would change the explanation altogether:
If cases appear in which the pay of the layer the tool ate actually rises, the split should be seen by the logic of technique, not the logic of layers.
In both, the question carries a way of seeing, that reviewing is a matter of how far to open the bundle, that work is divided into layers a tool can eat, and the test line names an observation under which that way of seeing should be given up for another.
The remaining six lines say what to observe without saying which result would tell against the question. Some promise that the observation will reveal or verify something:
It is verified by seeing whether the reply click rate per message and the subscription renewal rate fall together over the six months after obligatory sending is introduced.
This line is the most measurable in the selection, with two rates and a period, and it says plainly which joint fall would bear out the worry about obligatory messages. Others assert what a positive result would mean:
If the return-attendance rate rises at venues that have tried stating the share of prerecorded sound in their announcements, it is the statement, not deception, that brings the audience back.
These lines can be confirmed but are silent on disconfirmation; a flat rate of return, in the second case, is left without a reading. The difference matters for the persona debate. Chapman and Milham's complaint was that a persona cannot be verified or falsified. The role in these records cannot be either: no observation could show that a regular in the audience is the wrong figure to ask. But seven of the thirteen questions can be shown to rest on too large an assumption, by an observation the record itself describes. On those seven, the objection does not hold for the question, though it still holds for the role.
The synthetic status cuts in an unexpected direction here. Hainmueller, Hangartner and Yamamoto checked hypothetical answers against real choices. The Dionyas records invert the arrangement: the scene is avowedly invented, but every test line points at something that could be observed in the world, the silence of a room, the words of listeners, a rate of renewal, the pay of a layer of work. The invented part is the occasion; the test is aimed outward. Nothing in the records says that any of these tests has been run, and each is marked as registered and awaiting judgment.
1.4Who would run the test, and where the test comes from
A test needs someone to run it. The thirteen test lines are written in an impersonal voice, as instructions without a subject or as conditionals about what would follow. None names who would observe, compare or decide; the count is 0 of 13. The role at the head of the record is the natural candidate, but the records never say that the cover director runs the cover test or that the new reviewer plays the demo tracks to two people. Two lines name a meeting as the place where the result should land:
Set the words of fans who have heard the interlude at a show beside the words spoken where the recording is heard. If the two experiences differ, the next release meeting must hear that difference.
Put the condition backstage as one cell of the meeting's agenda, and the results of releases that were postponed and releases that were not can be compared. If the comparison yields a value, it means the body too has become grounds.
The meeting is an institution, not a person, and it appears here as the one that must hear, not the one that observes. The question of who holds the test, and who would say that a premise has failed, is left open by the records.
The form of the test, by contrast, is not open. It follows the series of the source card. All six records of series C have a test line of two sentences that prescribes a procedure of listening, asking or setting words side by side, and five of the six name a result that would weaken the premise. Neither record of series A does either; both name a count, a rate or a period and end by saying that the observation reveals or verifies. Series B falls between: one of its five lines has two sentences, two name a result that would change the explanation, one names a rate. The subjects do not sort this way. Series C asks about covers, choruses, choreography, nicknames, interludes and an outside composer's demo; series B about demo tracks made with AI, tools, venues, a director's ear and a release date; series A about lyrics and messages. Demo tracks appear in both series B and C, and the two records give them different kinds of test. So do the two records on live shows, which also sit in different series. The three questions stated as scenes rather than questions come one from series B and two from series C.
The records' own closing statement gives the reason this study can see that pattern, and the limit on what it can say about it. Each record says its source card was
a model's draft checked and registered by RISCON
and gives the card a date and a number. It does not say which model drafted which series, what the checking looked for, or whether a test line was among the things checked. This study therefore reads the pattern only as what it is: a regularity in the published records that runs with the card series and not with the subject. It does not attribute it to a model, a person or a method.
Timnit Gebru, Jamie Morgenstern and others, Datasheets for Datasets (2021), ask that a dataset come with a sheet documenting its motivation, composition, collection process and recommended uses. Read through that lens, that closing statement is a datasheet of one clause. It documents composition (the people, organizations, figures and events are set for the question), part of the collection process (a draft, checked, registered, on a date), and an implicit recommended use (not as a statement of actual events). It omits what would explain the regularity just described: the instructions behind each series and the criteria of the check. The lens fits badly in one respect, since a single inquiry is not a dataset; it fits well in another, since the thirteen together are a small collection whose differences can only be read through their making.
What the chapter adds to the two debates can now be put plainly. To the debate on vignettes it adds a case of the scene reduced to a role and a sentence, where the comparison a vignette study builds into its scenes moves into a test the record asks someone else to run. To the debate on personas it adds a way past the objection that an invented figure cannot be falsified: the figure still cannot be, but in seven of these thirteen records the question it carries can be, by an observation the record names. And to the call for documenting how records are made, it adds an example of how much a single clause of provenance can show, a regularity by series, and how much it leaves unexplained.