Discussion
The ten scenes answer the question this study put to them in a consistent way. What decides whether a loss counts as a misstep is, in most of them, a record: a line in a plan, a sheet of unfinished visits, a status that merges two meanings, a file that holds a date. In six of them the slip is placed in the seam where two records that were supposed to agree do not. Where a correction is proposed, it is aimed at the record that names the next loss, and the work that follows the slip is divided so that it does not stay with the one who found or admitted it.
This answer sits awkwardly in the debate it was read beside, and the awkwardness is useful. Reason, Human Error: Models and Management (2000), offers two places to look, the person on the front line and the upstream system. The scenes look at neither in full. They keep the person as one reading among three, and four of them keep a reading that names an error of a hand. Yet they rarely let that reading decide. Nor do they dissolve the slip into a whole system: the first coil does not blame the steelworks, and the empty doorstep does not blame the gas network. They point instead to one seam at a time, a plan and a mill, a timetable and a day, a screen and a room. This study reads that as a third place to look, smaller than the system and larger than the hand. It is a reading of written scenes, not a claim about where errors are in real work.
The scenes also show why it matters which record does the naming. Hollnagel (2018) observes that safety has been measured by counting what fails. The first coil makes the same observation from inside a single report: the coil lost is written down and the two coils kept are not. A count of talk about missteps is in this sense a count of what was named as failure. The scenes do not say such a count is wrong. They show, one written case at a time, how much of what it counts depends on which records were in place to call a loss a mistake.
The workaround literature fits the makeshift rest and the swapped container closely, and the fit has limits. Debono and colleagues (2013) conclude that workarounds enable, yet potentially compromise, the work they serve; the scenes write both halves into their readings and decline to choose before a check. Alter (2014) points to the route by which workarounds sometimes become planned changes. The makeshift rest and the pencilled drawing are scenes of that route stopped at its first step, the moment when a fix on the floor either reaches the drawing or does not. What the scenes add is attention to that step as a separate piece of work with its own time.
The fire lookout fits worst, and is the most instructive for that reason. Dixon, Wickens and McCarley (2007) found that false alarms from automation hurt overall performance more than misses. The scene's false alarm comes from a person, and its judgment treats the reproach that follows, not the false alarm, as the costly event. The two do not contradict each other; they look at different links. The experiment asks what false alarms do to the one who receives them. The scene asks what a word to the one who raised the alarm does to the next alarm. The scene states that it cannot say how much the delay it describes would cost in a real fire, and this study does not say either.
Contributions
The study makes three contributions. First, it shows that GAEKJU's synthetic scenes of missteps let a record, not the size of a loss, decide what counts as an error, and that six of the ten place the slip between two records, marks, times or appointments set against each other. Second, it shows that half of the scenes read a slip, a workaround or a wasted trip as news, and that half of the judgments decline to end in a person's fault or put off both praise and blame until a check is made. Third, it shows that the scenes ask a correction to reach a place that can keep it, and that most proposals divide the work after the slip so that the one who found or admitted it does not carry it alone. Read together, the three suggest a location for error between two records, which the person approach and the system approach in the debate on human error do not name as such.
Limits
The limits are those of the material. The scenes are synthetic; they are written for comparison and are not testimony from any trade, and nothing in this study describes what happens in steelworks, hotels or pharmacies. The selection rule keeps only the scenes whose evidence holds no digit, ten of 56; the scenes it sets aside may frame missteps differently, and this study does not speak for them. The counts are tests for Korean words chosen by this study, listed beside each count; another set of words would give other counts. The English is this study's translation, and a reading that turns on a single word may turn differently in the Korean. Finally, the works are read only through their notes: where this study goes past a note, it says so, and those readings are its own.
Further research
Three lines of work could follow. One would read the scenes the rule set aside, to see whether scenes that bring counts into their evidence place the misstep differently. A second would ask what a record would have to hold for the seam between two records to show in a count of talk about missteps. A third would take the scenes' own counter-conditions at their word: each names a record that would show its judgment wrong, a radio log, a thickness sheet, a list of uninspected homes before and after a change of hours. Whether any such record exists in a real workplace, and who would read it, is a question the scenes leave open, and this study leaves it open too.