1.1Why a literature review in a book of records
A book of records can be read in two ways. It can be read as it stands, with its own words and its own limits, or it can be read through ideas that were built elsewhere and for other purposes. This study does both, and it is worth saying at the start why the second is needed at all, and what it will and will not be allowed to do.
WSTiger's fifty-three decision records are synthetic settings published for study. Each holds a scenario, a question, a set of options, an example judgment, and a line saying when the judgment should be reassessed. None of them reports a real firm, a real trade or a real outcome. Read on their own, they show how a decision about risk can be written down before anyone knows how it turns out: what the decision rests on, what it gives up, and what would make it change. Read that way, however, they also leave almost everything else to the reader. A record does not tell the reader whether it belongs in a wider literature on probability, on organisational failure, on the value of waiting, or on how decisions are judged afterwards. Without some way of locating each record among the ideas that already exist, a reader has a set of interesting cases and no map.
Ideas, in turn, come with their own disagreements. The works that this study may cite are not a settled body of doctrine that any record can simply confirm or refute. Some of them are histories of how a concept came about. Some are arguments about what a concept should mean. Some are reports of experiments, of organisations under pressure, or of long runs of forecasts. They differ in what they study, in what counts as evidence for them, and in what they claim. Two works can use the same word — uncertainty, risk, deviation, reliability — and mean different things by it. A reader who picks up a record and applies the nearest idea without knowing the terms of that idea's own debate will usually read the record as an illustration of the idea rather than as evidence about anything.
That risk is worth stating in plain terms, because it is the failure mode of any book that pairs cases with theory. An idea drawn from a work has a scope: a claim about insurance markets does not settle a question about who holds a key, and a claim about accidents in tightly coupled systems does not settle a question about how a position is sized. A reader who knows the scope can ask of a record whether the idea reaches it at all. A reader who does not know the scope will find an illustration wherever the words are close enough together.
The works differ in three further ways that matter for how they are used here. They differ in what they study: an individual choosing under odds, a group running a hazardous system, a firm committing capital, a body judging a decision after it. They differ in what counts as evidence for them: a philosophical argument, a reconstruction of an incident, a formal economic argument, a series of scored forecasts. And they differ in what they are for. Some describe how people do decide; some argue for how they should; some are written to guide practice and some to settle a scholarly dispute. A book that uses them as lenses has to keep those purposes apart, because a description of behaviour is not a recommendation, and a recommendation is not a finding about what happens.
This chapter keeps them apart by writing the same small unit for each work: who holds the position, what the position says, and on what grounds. It does not argue with any of the works. Where two works disagree, this chapter reports the disagreement as the reading notes give it, and marks the places where the disagreement is one this study sets up rather than one the works themselves state.
This chapter therefore does one thing before the records are opened: it maps four literatures, in the brief form that the later chapters need. The first is on risk and uncertainty, and it asks what can be given a probability at all. The second is on organisations and accidents, and it asks why structures built to prevent failure produce failure. The third is on waiting and irreversibility, and it asks what it costs to decide now and what it costs to decide later. The fourth is on judging decisions and the people who make them, and it asks what an outcome can and cannot show about the decision that preceded it. A fifth, shorter strand — the history of probability itself — runs under all four, because each of them assumes that some decisions can be weighed in numbers and others cannot.
Each of these literatures gets its own chapter later, where the records are read against it in full. What this chapter does is deliberately smaller. It states who holds which position, on what grounds, in a paragraph each, so that when a later chapter says that a record bears on a debate or cannot reach it, the reader knows what the debate is. Nothing in this chapter is tested against a record, and no case is read in it. The comparison with cases begins in the chapters that follow.
One rule applies to everything in this chapter and in the chapters after it. The literatures are lenses, not verdicts. Where a work fits a case badly, this study says so. Where the records bear on a debate, this study says where they bear and where they cannot reach. The method that makes records comparable at all — the decision ledger, its axes, its two coders and its pre-registered hypotheses — is set out in Chapter 2, which is the chapter that follows this one and which the later chapters depend on.
1.2How probability became a way of deciding
Before any of the four literatures can be read, one background has to be set: the idea that uncertainty can be made into a number that a decision can be built on. Some of the works reviewed below build on that idea, and others are arguments against it.
Ian Hacking, in The Emergence of Probability (1975), places the emergence of probability as a concept around 1660. Hacking's reading note gives the concept two faces from the start. One face is the frequency of chance events: how often a thing happens among instances of it. The other face is the degree of belief warranted by evidence: how far a given body of support reaches. These are not the same thing, and they do not always move together. An event can be frequent in the first sense while the evidence warranting any belief about it is thin, and a belief can be well supported in the second sense even though the event it concerns is rare. This study reads Hacking's point as being that the two have been joined from the beginning, and that the join has never been a matter of course.
Lorraine Daston, in Classical Probability in the Enlightenment (1988), follows probability into the eighteenth century as a model of reasonable judgment. Her reading note specifies that this includes its use in insurance and annuities. This study reads those as uses in which a decision must be made before the answer comes. Probability, on Daston's account, is not only a description of how often things happen; it is an argument about what a reasonable person may conclude from what they know. This study reads Daston as giving probability a second job beyond counting — that of authorising a decision in the absence of a known outcome — which is the job the later literatures argue about.
This study reads that use as the oldest form of the problem its own subject holds: a decision is taken now, on the strength of a class of cases, about an individual case whose outcome no one knows. The two faces Hacking describes meet in it, because the class is what can be counted and the individual case is what must be believed in.
Peter L. Bernstein, in Against the Gods: The Remarkable Story of Risk (1996), offers a popular history of the idea of risk, running from the arithmetic of chance to modern finance. For this study it is useful for a different reason: it traces the direction in which the idea travelled, from the arithmetic of chance into the modern finance that the records in this study describe. This study reads Bernstein as showing that the number is not a natural fact about a decision but something that grew up around decisions and then became part of how they were taken.
There is a limit to what this background can carry, and the limit is worth marking before the four literatures are opened. A history of a concept tells the reader when the concept arrived and what it was made of; it does not tell the reader what any particular decision should rest on. Bernstein's history runs into modern finance, and modern finance is the setting of the records this study reads, but the direction of travel is no warrant for any practice in that setting. This study reads the three works as establishing a background only: that the weighing of uncertain decisions in numbers has a history, that the number entered practical business early, and that it has carried more than one meaning since it arrived.
The background also explains why the four literatures that follow argue past one another as often as they argue with one another. A writer whose object is a frequency of events and a writer whose object is the reasonableness of a belief are not disagreeing about a fact; they are using two faces of the same concept. Hacking's note gives the reader the vocabulary to see this. Daston's note shows the concept doing practical work in the eighteenth century. Bernstein's note shows how far the idea travelled afterwards. None of the three decides any of the arguments that follow in this chapter, and this chapter does not use them as if they did.
The chapters that come after carry this background with them. First, the assumption that a decision can be given a number is itself a claim with a history, not a neutral starting point. Second, the two faces Hacking describes give later writers two different objects to fight over: frequency, which needs a class of like cases, and warranted belief, which does not. Third, the practical uses Daston names — insurance, annuities — are precisely the uses where a decision is taken before its outcome and where the question of who waits and who pays is unavoidable. That last point is the hinge between this background and the third literature in Section 1.5.
This study keeps these three works at the level of their reading notes, and does not claim for them anything beyond them. Where the study draws its own inference — for instance, that Daston's account shows probability doing the work of authorising a decision as well as describing a frequency — that inference is this study's, and it is marked as such wherever it is used.
1.3First literature: risk and uncertainty
The first literature asks a question that sounds simple and is not: what can be measured, and what cannot. Its central figure is Frank H. Knight, and the position it is still arguing about is his.
Knight, in Risk, Uncertainty and Profit (1921), separates measurable uncertainty from unmeasurable uncertainty. Chapter VII makes the separation; Chapter VIII, titled "Structures and Methods for Meeting Uncertainty", names the first kind risk and the second kind uncertainty. His own sentence in that chapter sets the terminology: "To preserve the distinction which has been drawn in the last chapter between the measurable uncertainty and an unmeasurable one we may use the term 'risk' to designate the former and the term 'uncertainty' for the latter."
The consequences Knight draws are not symmetrical. Insurance and grouping deal with risk: many instances of a class are pooled, and the individual case is treated as an instance of it. True uncertainty, on Knight's account, cannot be insured. Profit, in his argument, is tied to bearing that uncertainty; this study reads that bearing as judgment that cannot be delegated to a formula or to a market. This is the position that the later readings of Knight in this section argue over.
John Maynard Keynes, in A Treatise on Probability (1921), offers a different account in the same year. For Keynes, probability is a logical relation between evidence and conclusion, and it is not always numerical. His reading note also gives him a second term that is central to this study's subject: the weight of evidence differs from the probability itself. Two decisions can be given the same probability while resting on very different amounts of support, and Keynes's distinction says that the difference is real and not reducible to the number. This study reads Keynes as providing the vocabulary that the records will later be read with: a probability is not the whole of what a decision rests on.
Daniel Ellsberg, in Risk, Ambiguity, and the Savage Axioms (1961), gives Knight's distinction an experimental form. His reading note records the pattern: people prefer a bet whose odds are known to an otherwise equal bet whose odds are unknown — a preference he calls ambiguity aversion, which breaks the axioms of subjective expected utility that Savage had set out. Ellsberg does not settle what the distinction means; he shows that people act as if it means something. For this study the relevance is that ambiguity aversion is a pattern of choice when the odds themselves are unknown, even though a person might still propose a probability of the outcome.
John Kay and Mervyn King, in Radical Uncertainty: Decision-Making Beyond the Numbers (2020), carry the argument into the present. Their reading note holds that many decisions face radical uncertainty that probabilities cannot capture, and that in such cases the useful question is not a numerical one but the question of what is going on here. They recommend reasoning with narratives and choosing strategies that are robust and resilient to what cannot be foreseen rather than optimising on invented numbers. Their position is later than the others and speaks more directly to practice, which makes it useful for reading records that are themselves written as practice.
Two later readings of Knight's distinction divide the first literature. Stephen F. LeRoy and Larry D. Singell Jr., in Knight on Risk and Uncertainty (1987), argue that Knight's distinction does not turn on whether people can form probabilities — they hold that Knight took subjective probabilities for granted — but on whether insurance markets can exist. On their reading, uncertainty is where moral hazard and adverse selection keep a risk from being insured. Richard N. Langlois and Metin M. Cosgel, in Frank Knight on Risk, Uncertainty, and the Firm: A New Interpretation (1993), answer them: for Knight, uncertainty arises where instances cannot be sorted into classes of like cases, so no frequency can be had, and such situations call for judgment, with the entrepreneur's judgment as what the firm is built around. The two readings differ on where the difficulty lies — in the failure of a market to form, or in the absence of a class to count — and the difference matters for this study, since a record may or may not offer a class of like cases to count. Chapter 3 sets the two readings out fully and reads the records that sit at the edge of the measurable against both.
What the first literature gives the rest of the study is a set of distinctions that the records can be read with. The first is the one Knight names: a matter that can be measured and a matter that cannot, and the fact that one action can contain both. The second is Keynes's weight of evidence: two decisions can rest on the same probability and different amounts of support. The third is the disagreement between the two readings of Knight, which places the difficulty either in a market that will not form or in a class of cases that cannot be counted. The fourth is the answer of Kay and King, which asks what is going on here before asking for a number at all.
These distinctions are not interchangeable, and the temptation to slide between them is strong. A difficulty that comes from having no class of like cases is not the same as a difficulty that comes from a market failing to form, and neither is the same as a difficulty in measuring an outcome that could in principle be counted. This study keeps them apart by saying, when a record is read against the first literature, which of the distinctions the record reaches and which it does not.
None of these distinctions is a verdict on any record, and none of them can be applied by resemblance alone. A record whose question is about the size of a position, or about the scope of an automated system, may have nothing to do with the measurability of an uncertain event, and the honest answer is that the first literature does not reach it. That kind of answer is common enough in the later chapters to be worth preparing for: a literature that fits badly is not thereby refuted, and a record that it cannot reach is not thereby without interest.
1.4Second literature: organisations and accidents
The second literature starts from a different scene. It does not begin with a decision made by an individual facing a number; it begins with a system in which many people, procedures and pieces of equipment are already in motion, and it asks why systems built to prevent failure produce failure anyway.
Barry A. Turner, in Man-Made Disasters (1978), gives the process a name and a shape. His reading note describes disasters as having an incubation period in which unnoticed or misread events — events at odds with the accepted beliefs about a hazard — accumulate, until a precipitating event brings them together. Turner's claim is about the direction of causation: disasters are failures of foresight produced by organisations, not acts of fate. This study reads the incubation period as the interval its subject concerns most directly, because during it events accumulate — some unnoticed, some misread — while the organisation's course does not change.
Charles Perrow, in Normal Accidents: Living with High-Risk Technologies (1984), makes a structural argument. His reading note gives him two independent dimensions: interactive complexity and tight coupling. Systems high on both are, in his phrase, ones in which accidents are normal — to be expected from the structure rather than from a single error. This study reads Perrow's position as leaving little room for remedies that address one failure at a time, since in a system of this shape the structure, not the single error, is what produces the accident.
James Reason, in Managing the Risks of Organizational Accidents (1997), works at the level of the organisation's own anatomy. His reading note gives him two contributions that later chapters return to: latent conditions in an organisation line up with active failures, and defences in depth have holes. Reason's picture is of defences that are real but imperfect, and of preconditions laid down long before any event that make an active failure consequential when it finally occurs.
Diane Vaughan, in The Challenger Launch Decision (1996), names the process by which an organisation comes to accept as normal a departure it once treated as a warning — the normalisation of deviance. Her reading note presents it as a course an organisation follows over time rather than as a single lapse. This study reads the normalisation of deviance as one way an organisation's reading of a departure changes over time, close to Turner's incubation but not identical with it: Vaughan's process concerns a departure the organisation once treated as a warning, whereas Turner's accumulating events may equally go unnoticed.
Against this body of argument stands the high-reliability school. Todd R. La Porte and Paula M. Consolini, in Working in Practice but Not in Theory: Theoretical Challenges of "High-Reliability Organizations" (1991), record a finding that existing organisation theory struggles with: some organisations — their note names air traffic control and aircraft carrier flight operations — run hazardous work with very few failures. Organisation theory, which expects error and trial-and-error learning, has difficulty explaining how this is done. Karl E. Weick and Kathleen M. Sutcliffe, in Managing the Unexpected (2001), set out the practices that belong to organisations of this kind: they attend to failure, refuse to simplify, stay close to operations, keep the capacity to recover, and defer to expertise where it is. Weick's account describes how these organisations actively produce reliability.
Scott D. Sagan, in The Limits of Safety: Organizations, Accidents, and Nuclear Weapons (1993), tests the two positions against a real record. His reading note states that he sets normal accidents theory against high reliability theory and tests them on the command and control of United States nuclear weapons, including incidents during the Cuban missile crisis, and that he finds the record closer to normal accidents theory, with dangerous near-accidents and with organisational learning limited by secrecy and blame. This study reads Sagan's contribution as methodological as well as substantive: the choice between the two theories cannot be settled by looking only at a system that has behaved well so far, when the record he examines holds dangerous near-accidents.
This study does not resolve the debate between normal accidents and high reliability. Chapter 5 treats both positions in full and reads records against them — a threshold, a limit raised one time at a time, a scope that widens with success, and a stop that has never been rehearsed — and says plainly where synthetic records with no outcomes cannot reach. What the records can show, and what this chapter prepares the reader to look for, is whether a record writes the accumulating events, who reads them as warnings, and who could have changed the course while the interval was still open.
The second literature also differs from the first in the shape of its evidence and in the shape of its disagreement. Its objects are incidents and organisations rather than odds: the command and control of nuclear weapons tested against two theories, operations such as air traffic control and aircraft carrier flight operations run with very few failures, an organisation's treatment of a departure over time, and the accumulation of unnoticed or misread events before a disaster. The disagreement is not about what a number means but about what structures produce. Perrow's two dimensions make accidents a property of a system's shape. The high-reliability school makes reliability a property of what an organisation does. Reason's latent conditions are preconditions laid down before any event, and his defences in depth are barriers with holes. Vaughan's normalisation of deviance is a course an organisation follows over time.
A reader of records can hold these apart by asking a different question of each. Perrow asks what the system is like. La Porte and Consolini, and Weick and Sutcliffe, ask what the organisation does. Reason asks what was already in place. Turner asks what accumulated before the event. Vaughan asks how the reading of a warning changed. Sagan asks what a record of near-accidents shows about which of these accounts fits. Those questions are not answered by the same evidence, and a record that answers one of them says nothing about the rest.
1.5Third literature: waiting and irreversibility
The third literature is the one closest to this study's own question, because its subject is the interval itself. It asks what it costs to decide now, what it costs to decide later, and what happens to the alternatives once a decision has been made.
Kenneth J. Arrow and Anthony C. Fisher, in Environmental Preservation, Uncertainty, and Irreversibility (1974), make the case that waiting is not merely a delay in an answer but a decision with a value. Their reading note holds that when a development cannot be reversed and its future benefits are uncertain, the information that will arrive later gives value to keeping the option to preserve — value later called quasi-option value. Ignoring that value, on their argument, biases decisions towards irreversible commitment. The subject of their note is environmental preservation. This study reads the structure of the argument as extending past that subject, and as resting on the conditions the note names: the commitment cannot be reversed, the future benefits are uncertain, and information bearing on them will arrive later while the option to preserve is still open.
Avinash K. Dixit and Robert S. Pindyck, in Investment under Uncertainty (1994), carry the same structure into investment. Their reading note holds that an investment that cannot be undone and can be delayed carries an option value of waiting, and that the net present value rule ignores it, so that it can be rational to wait for information even when investing now looks profitable. The two works together give the third literature its standard position: where a commitment cannot be undone and can be delayed, and what later information will show is uncertain, waiting carries an option value that the net present value rule, which discounts future returns but leaves that option out, will miss.
David Collingridge, in The Social Control of Technology (1980), adds the reason why such waiting is so hard to organise. His reading note names the dilemma of control: while a technology is young its effects cannot be known, and once they are known it is entrenched and hard to change. Collingridge's answer, in the same note, is to keep decisions corrigible — easy to monitor and to correct — and to keep options open. This is a different prescription from waiting for a decisive moment, and the difference is the point: it argues for structures that can be revised while the answer is still not in, rather than for a single judgment made at the right time.
Charles E. Lindblom, in The Science of "Muddling Through" (1959), sets a different picture of decision-making beside the other three in this section. His reading note holds that administrators in practice decide by successive limited comparisons of a few options that differ little from present policy, not by comprehensive rational analysis, and that Lindblom defends this incremental method as the one that works. Lindblom's position shifts the emphasis from one moment of choice to a series of limited comparisons with present policy. This study reads that shift as a caution for any method that treats a decision as a single moment of commitment followed, if things go wrong, by reversal.
These four positions do not sit comfortably together, and the tension between them runs through what this study later does with the records. The records themselves are written at the moment before the answer, which is precisely the moment both Arrow and Fisher's argument and Lindblom's description care about. What the records can bear on is narrower than the theories: they show, in the way they are written, whether delay is presented as an option with a value, whether a decision is presented as reversible, and whether the record even names what could still be changed.
The third literature has a further feature that sets it apart from the two before it: it is about a choice between acting and not acting, where not acting is itself a choice with consequences. Arrow and Fisher's argument attaches value to keeping an option open; Dixit and Pindyck attach an option value to waiting; Collingridge asks for structures that can be corrected once effects show themselves; Lindblom describes decision-making as a run of small comparisons with present policy. Read together, they make the moment of decision less like a threshold and more like a position from which the next move is chosen.
That is also why they are hard to apply to a record. A record names a question and its options at one moment, and it says what would make the judgment change. The theories state the value of waiting in general terms, and the readings of the records ask whether waiting appears as an option at all.
1.6Fourth literature: judging decisions and their makers
The fourth literature turns the question around. Instead of asking what a decision rests on while it is being made, it asks what can be learned from a decision once it has been made, and by whom.
Amos Tversky and Daniel Kahneman, in Judgment under Uncertainty: Heuristics and Biases (1974), report that people judge probabilities with a few heuristics — representativeness, availability, and anchoring and adjustment — which usually serve but lead to systematic and predictable errors. Their reading note is careful to call these systematic rather than random. Daniel Kahneman and Amos Tversky, in Prospect Theory: An Analysis of Decision under Risk (1979), turn to choice rather than judgment. Their reading note gives the pattern: people judge outcomes as gains and losses from a reference point, losses weigh more than gains of the same size, and probabilities are weighted rather than taken as they are. This study reads the two works together as describing a decider who does not calculate but compares, and whose comparisons turn on a reference point and on a few heuristics.
The next two works concern how a decision is evaluated after it is over. Baruch Fischhoff, in Hindsight ≠ Foresight: The Effect of Outcome Knowledge on Judgment under Uncertainty (1975), shows that people told how events turned out judge that outcome more likely than they would have without knowing it, and do not realise that the knowledge changed their judgment; Fischhoff calls this creeping determinism. Jonathan Baron and John C. Hershey, in Outcome Bias in Decision Evaluation (1988), add a related pattern: people rate the same decision, made on the same information, as better when its outcome was good than when its outcome was bad. The two effects together mean that a decision evaluated after its outcome is not evaluated on the same terms as the decision itself.
Two further works ask how a judgment can be tested once the outcome is in. Glenn W. Brier, in Verification of Forecasts Expressed in Terms of Probability (1950), proposes a score for probability forecasts — the mean squared difference between the forecast probabilities and what happened — and his reading note adds the point that a forecaster cannot improve on this score by stating anything but the probability believed. Philip E. Tetlock, in Expert Political Judgment (2005), studies expert forecasts over a long period. His reading note records that experts were poorly calibrated and often no better than simple rules, and that those who drew on many ideas and revised readily — foxes — did better than those committed to one big idea — hedgehogs. On this account, what is checked is the gap between the probability a forecaster stated and what happened, scored over a run of forecasts.
Two further works concern who carries the risk in the meantime, and this study's fourth literature is deliberately extended to include them because the question of the bearer cannot be separated from the question of judgment. Niklas Luhmann, in Risk: A Sociological Theory (1993), separates risk from danger: risk is a possible loss attributed to one's own decision, danger is a possible loss attributed to something outside one's control, and the same loss is a risk to the one who decides and a danger to those who are affected but did not decide. Michael Power's two books trace what organisations do with uncertainty. In Organized Uncertainty: Designing a World of Risk Management (2007), risk management is a way organisations turn uncertainty into objects they can manage and account for. In The Risk Management of Everything: Rethinking the Politics of Uncertainty (2004), risk management spreads into all organisational activity and turns towards secondary risks — the reputational and legal risks to the organisation and its managers — so that defensive documentation can take the place of judgment.
Chapter 7 reads the heuristics, prospect theory, hindsight and outcome bias against records that keep the reason before the outcome, and against Brier and Tetlock, where the records are read as evidence about how a decision is judged over a run of decisions. Chapter 10 reads Luhmann's distinction against the records on who waits and who pays, and Chapter 15 sets Power's two books beside the ledger's rows on the bearer of the waiting and on what is written down as grounds. None of these chapters claims that the records settle any of the positions; the records have no outcomes, and part of the fourth literature — the verification of forecasts — needs exactly that.
The fourth literature is also the one that comes closest to this study's subject, because its subject is the distance between a decision and its answer. Its works divide along that distance. On the near side are the works on how a decision is made when the answer is not in: a reference point, a few heuristics, and weights that are not the probabilities themselves. On the far side are the works on what the answer does to the judgment of the decision: hindsight, outcome bias, and the scoring of a forecast against what happened. And beside both stands the question of who carries the risk while the answer is pending.
This study reads those three strands as one problem with three parts: what the decider had, what the answer later does to the reading of it, and who paid in between. The three parts are easy to collapse into one another, because all three can be called evaluation. Keeping them apart is what allows one record to be compared with another on the same terms.
1.7The gap
The four literatures and the background behind them have one subject in common, and this is the point at which the narrower subject of this study can be stated. Each of them studies something other than a written decision record. That is not a criticism of any of them, but it does mark where this study has to place itself.
The first literature studies probability, its history and its uses. Its evidence is philosophical argument, the history of a concept, and, in Ellsberg's case, a pattern of choice shown on hypothetical bets with known and unknown odds. The second literature studies organisations under pressure: accidents, near-accidents, launches, warnings, and the practices of a small number of highly reliable operations. Its evidence is the reconstruction of events and the theory of organisational structure. The third literature studies waiting and irreversibility, and much of its argument is economic: the value of information that will arrive later, and the structure of a decision that cannot be undone. Collingridge within it studies the control of technology, and Lindblom administrative decision-making. The fourth literature studies judgment, forecast and evaluation, and its evidence runs from hypothetical choice problems to long series of forecasts, with Luhmann's distinction and Power's organisational studies beside the experimental work. This study reads the gap narrowly: on these reading notes, none of these literatures takes as its object a decision set out before its answer, set beside other such decisions on the same questions, and read for what it says about the interval.
That is the gap this study occupies, and it is deliberately narrow. The fifty-three records in this study are not new evidence about markets, about organisations or about how people choose. They are a set of synthetic settings published by WSTiger for study, and this study reads them case after case on the same six axes — who decides, on what grounds and with what missing, by when and who waits and pays meanwhile, whether the decision can be undone, what would make it change, and when it was judged. The comparison is made possible by a method rather than by a theory: a decision ledger that codes every record the same way, read by two models of different families, with the agreement between them reported, with strict and lenient readings, and with pre-registered hypotheses whose verdicts come from counts. That method is set out in Chapter 2, and every later chapter depends on it.
What this study does not add is worth stating in the same breath. It contributes no new theory of uncertainty. It does not extend, correct or resolve the debate between LeRoy and Singell and Langlois and Cosgel over what Knight meant; it reads records against both readings and reports what the records can and cannot show. It contributes no new evidence about real markets, real firms or real trades: no record in this study reports one, and none of them has an outcome. It adds nothing to the arguments of Turner, Perrow, Reason or Vaughan about organisational failure, nor to the argument between normal accidents theory and high reliability theory, nor to Arrow and Fisher or Dixit and Pindyck on the option value of waiting, nor to Fischhoff and Baron and Hershey on the distortion of judgment by outcome knowledge. Where the records bear on those arguments, this study says so; where they cannot reach, it says that too.
The restraint has a practical reason as well as a scholarly one. A synthetic setting cannot establish what happens in a market or an organisation. Theories tell the reader what to look for; the records show what these particular settings write down. Keeping the two apart is what makes the comparison in this study possible, and it is also what keeps a reader from treating a well-known idea as proved by a scene that merely resembles it.
What the records can bear on is a question those literatures raise only at the edges: how the time between a question and its answer is written down. A decision record is unusual among the objects these literatures study, because it sets a decision out before its resolution and asks what the decision rests on at that point. That means it carries, in its own form, the reasons given at the time, the options that were open, and the condition under which the decision would be reconsidered. Whether it carries anything about the deadline that was absent, the cost of waiting, or the person who would bear that cost is an empirical question about the records, and it is the question the study answers with counts rather than with argument. Chapters 7 to 13 read the records on those axes and set what they show beside the literatures mapped here.
The reader who arrives at Chapter 2 with this map in hand will find that the map is used sparingly and in one direction. Each later chapter treats its own ideas in the depth its subject needs, and this chapter's brief statements make no claim to be their full treatment. The chapter that follows sets out the method that makes the comparison possible at all, and after it the records begin.