You Should've Known

I. The life you were supposed to have lived

A capable researcher enters an unfamiliar professional group, raises a question publicly, and is told that such questions go through a private conversation first. Someone adds that, with her experience, she should have known.

The rule might protect something important. The reference to her experience already contains a theory of her life: people like her encounter certain situations, recognize what matters, and acquire the relevant judgment. Her failure to know violates that prediction.

“The train should have arrived” expresses an expectation. “You should have checked the timetable” assigns a responsibility. “You should have known” can slide between the two without making the transition explicit.

Its charitable reconstruction is: Given the history I thought you had, I expected you to have learned this. That is a useful observation. A surprising gap can reveal a missed lesson. It can also reveal that the observer imagined the wrong history.

An anomalous result identifies a problem in the prediction, not automatically a defect in the person predicted.

She might have learned her discipline where public questions were encouraged. She might have worked alone. She might have met the same rejection before without knowing whether it was a general rule or one person’s preference. Years of experience do not specify which encounters occurred, or what they taught.

The demand can conceal an entire alternative biography: you should have spent enough time in the environments that teach what this environment regards as elementary. That biography has costs. Time spent acquiring one community’s fluency is time not spent acquiring something else. A requirement to know a particular convention needs a reason that applies to the person’s actual responsibilities, rather than merely to the observer’s preferred route through life.

Surprise does not settle what the missing knowledge licenses us to conclude.

II. When thinking needs an observation

The claim that someone could have worked it out deserves the same scrutiny. “The information was available” leaves open how much work was required to find, combine, and apply it.

A consequence can follow from a set of premises without being cheap to derive. Knowing several mechanisms that could govern a social interaction does not reveal which will dominate this interaction. The missing element might be the local circumstances, the relative strengths of competing effects, or the recognition that a particular fact is relevant.

After an event, the outcome supplies a search key. Before it, there may have been many plausible chains to investigate.

Simulation exposes implications and possible failure modes. It leaves its own assumptions untested. Ten internally convincing accounts of how an audience might respond are still accounts whose relation to the actual audience needs checking.

A simulated reader asked to criticize an argument has already been assigned the task of reading it. The real obstacle might be obtaining anyone’s attention. Improving the simulated criticism can leave that obstacle untouched.

Sometimes an observation is a cheaper and more reliable computation than further private reasoning. A question reveals whether a rule applies. A small demonstration reveals a misunderstanding that neither side had anticipated. Someone else’s documented experience supplies what personal experimentation would otherwise have to discover.

Control theory calls an action dual when it both acts on an uncertain system and helps reveal how that system behaves. The idea is Feldbaum’s; later work studies the trade-off between immediate performance and information useful to subsequent decisions.¹

Ordinary limitations are enough. An agent has finite attention, an incomplete model, and alternatives for spending its next unit of effort. More thinking earns its place when there is a good reason to expect enough improvement in the decision to justify its cost, just as an experiment does.

The pertinent question is what the next operation adds. Does another simulation expose a dependency, use new evidence, or test an assumption? Does an actual attempt distinguish live possibilities? Does the remaining uncertainty matter to the decision at all?

A demand to “think first” is useful when thought avoids an unnecessary mistake. Used without a stopping condition, it can forbid the observation needed to finish thinking.

III. The risk that competence requires

Imagine two situations that are indistinguishable using the evidence affordable before an attempt. In one, an unfamiliar approach opens a valuable possibility. In the other, it fails. A bounded attempt would reveal something useful about which situation obtains.

A rule that forbids the attempt because it could fail also forgoes that information. Another source might supply it more cheaply. The possible discovery might not justify the risk. But where the attempt is the best available way to resolve a consequential uncertainty, a competent policy must accept its unfavorable branch.

From outside, that unfavorable branch may resemble an episode of poor judgment. The observer sees a failed approach, not the alternatives considered, the uncertainty remaining, or the value that would have followed from a different result. Competent exploration and careless action can generate the same visible episode.

The necessity is to permit justified inquiry without demanding a guarantee that it will look competent afterward.

Appearance is imperfect evidence for the process, not worthless. The comparison includes never trying: staying inside familiar tasks hides some gaps; expanding into unfamiliar ones reveals more of them. Curiosity does not confer unlimited claims on other people’s attention, permission to endanger them, or a right to repeat a settled experiment at their expense.

An informative attempt needs feedback that can reach the learner and bear on a later choice. “I was learning” is incomplete until it names the uncertainty, the observations that could have changed later decisions, and whether the result warranted revision, confirmation, or remained ambiguous. A requirement that every individual attempt look prudent after its outcome is known would reject part of the process that made later prudence possible.

IV. Knowing the rules of the group

Local fluency adds another ambiguity: understanding a rule, endorsing it, and deciding to comply with it are different achievements.

Alfred Schütz’s account of the stranger examines what happens when a newcomer cannot rely on the taken-for-granted interpretations through which established members navigate a group. What members use as familiar practical guidance becomes something the newcomer must investigate.²

That investigation has two possible directions. The newcomer can discover a constraint they missed. They can also discover that a convention performs a different function from its advertised one. Neither conclusion follows from being the newcomer.

Return to the researcher. A private conversation might prevent the needless exposure of confidential information. It might let a specialist correct a misunderstanding cheaply. It might instead let senior members stop inconvenient questions from becoming visible. The instruction “raise this privately” does not distinguish these mechanisms. Their consequences do.

There is also a case in which she knows the convention perfectly well and departs from it deliberately. Predicting an adverse reaction prices a cost. It does not settle the objective.

A low-probability attempt to find a collaborator can still be worthwhile. An objection that will not persuade the current decision-maker can nevertheless place relevant evidence before others. Whether either action makes sense depends on its expected effects and costs, including the attention it demands from people who did not volunteer to help.

“You would have known not to try” can therefore mean several things. The person lacked a fact. They rejected a premise. They valued a different outcome. Or they accepted a risk the speaker would not accept.

Treating all four as ignorance turns disagreement about objectives and assumptions into a test of belonging. The informed person becomes, by definition, the person who reaches the locally approved decision.

A convention’s familiarity supplies evidence about how people behave. Its justification still depends on what that behavior produces. Understanding the local equilibrium is useful precisely because it makes those consequences easier to distinguish.

V. How a rebuke changes the evidence

The outcome also changes the observer. In Baruch Fischhoff’s hindsight experiments, learning what happened altered judgments of its prior likelihood and of which preceding facts mattered. Participants did not fully recognize this influence.³

A related error concerns the decision rather than the forecast. Jonathan Baron and John Hershey found that people evaluated decisions more favorably following favorable outcomes, even when the information available to the decision-maker was held constant.⁴

These findings identify errors in evaluation, not reasons to ignore results. Outcomes supply new evidence about a model and, when other information is incomplete, about competence. A failed attempt can justify changing course. It cannot retroactively make its newly acquired evidence available to the earlier decision.

The social consequences extend beyond this retrospective mistake. Suppose a workplace penalizes conspicuous ignorance while providing little opportunity to distinguish responsible inquiry from negligence. Employees can protect their standing by avoiding unfamiliar tasks, suppressing questions, and hiding mistakes.

Each move changes what their colleagues can learn. Avoided tasks produce no observations. Unasked questions expose no misunderstandings. Hidden mistakes provide no shared warning. If the resulting silence is then interpreted as evidence that everyone understands, the workplace becomes more confident on the strength of information it has stopped collecting.

Jerker Denrell and James March model a related problem as the hot stove effect: an initially unfavorable experience reduces subsequent sampling, preventing some mistaken negative judgments from being corrected. Their model also retains the cost of counteracting this tendency: continued exploration sometimes prolongs genuinely bad choices.⁵

Social penalties can add another reason to stop sampling. The mechanism requires no conspiracy. A supervisor wants fewer errors, employees want to appear reliable, and both respond to visible failures. If neither examines what has disappeared from observation, apparent competence becomes easier to maintain than actual improvement.

Amy Edmondson’s hospital field study found differences between work groups in the detection and reporting of errors. Detected-error totals therefore had to be interpreted alongside how willing and able people were to expose mistakes. A more visible problem need not be a larger underlying problem.⁶

Over successive cohorts, this also changes what gets taught. “Nobody does that” passes from experienced members to newcomers. Where the warning reliably prevents waste, the inheritance is valuable. Where it preserves an obsolete or mistaken judgment, discouraging every test helps the judgment outlive its evidence.

The rebuke has become more than an assessment of someone’s past learning. It helps determine which learning will be possible next.

A group can demand evidence-based action while penalizing the actions through which missing evidence would become available.

VI. A requirement without an acquisition process

Some knowledge really is required. A person entrusted with equipment must know its relevant operating limits. An investigator must perform checks appropriate to the claims others will rely on. A collaborator cannot indefinitely make other people absorb the consequences of refusing to learn.

Discovering how ignorance arose tells us where to intervene. The requirement remains.

Suppose an operator misses a changed procedure. In one version, the change was clearly communicated, the operator had time to review it, and a required check was skipped. In another, management silently replaced the document, failed to provide access, and retained an assessment that tested the obsolete procedure. The same missing knowledge points to different repairs.

The 1987 investigation of the capsize of the ferry Herald of Free Enterprise is a documented case of that split. The assistant bosun, whose duty it was to close the bow doors, was asleep in his cabin when the ship left Zeebrugge. The formal investigation also found a company that, from top to bottom, was “infected with the disease of sloppiness,” and that had dismissed a request for a bridge indicator of door position as frivolous spending to detect employees failing at their jobs. Repeating that someone in the role must close the doors did not produce a process that closed them.⁷

Sanford Goldberg makes the underlying distinction explicitly in Should Have Known: legitimate expectations about someone’s knowledge can remain unmet without that person being culpably ignorant. The defect can lie in how a practice trains or equips its participants.⁸

A role can require a capability that the organization has failed to produce. Recognizing this need not lower the standard. It can require training, supervision, reassignment, or suspension of an activity whose prerequisites are absent.

Blaming the operator for the second version leaves the failed training arrangement intact. Excusing the first version as an inevitable consequence of everyone having different experiences leaves the neglected responsibility intact. Both responses avoid the causal question.

What process was supposed to produce the knowledge, and what prevented that process from succeeding?

The answer need not excuse anyone. Deliberately avoiding an inconvenient warning is itself part of the explanation. Nor must the process supply advance instruction for every imaginable contingency. Expertise includes recognizing limits, asking relevant questions, and withholding action when a missing answer makes the risk unacceptable.

“Someone in this role must know” cannot do the work of producing a person who knows. When the requirement survives and the acquisition process fails, repeating the requirement more forcefully does not repair the missing process.

VII. A “should” that improves judgment

A useful standard needs to produce reliable knowledge without making other people subsidize uncontrolled experimentation. That requires distinguishing failures of preparation from failures encountered during justified exploration—and examining the arrangements that produce both.

For an individual decision, the relevant comparison is a feasible alternative available beforehand. Which warning was neglected? Which check was worth its cost? Which responsibility made the missing knowledge necessary? General intelligence and the eventual existence of a convincing explanation do not answer those questions.

For a group, the comparison includes how it communicates requirements, permits questions, allocates practice, and handles the evidence of error. A newcomer should be able to discover the relevant rules without having to violate every one of them. Members should also have a way to challenge a rule without their disagreement being treated as proof that they never understood it.

The results must remain available to successors. A discovered limit becomes a warning others can find; a corrected convention stops being taught in its obsolete form. Shared knowledge spares people experiments that have ceased to be useful. Continued inquiry keeps that inheritance answerable to changed conditions.

Return once more to the public question. If it exposed confidential information despite a clear warning, the correction concerns a neglected duty. If the newcomer could not discover the convention, the group needs a better introduction. If the convention obstructs useful criticism, teaching it more efficiently reproduces the obstruction. If the attempt was proportionate and produced valuable information, its adverse reception need not condemn the decision.

These possibilities cannot be distinguished by the sentence “you should have known.” They require the facts it compresses.

The repair can be small. “That document contains the restriction you needed to check” identifies a missed step. “We never told new members how this works” identifies a communication failure. “Your prediction was reasonable, but this result changes the next attempt” preserves learning without inventing retrospective omniscience.

More elaborate procedures belong where consequences justify them. Turning every question into an application for permission would make the protection of inquiry another obstacle to inquiry. The same test applies to a proposed repair: does it improve preparation and correction, or merely make it easier to demonstrate that the prescribed steps were followed?

The learner has a corresponding responsibility to let the world correct the project. Accepting the risk of looking foolish is useful only while the process remains capable of discovering that an approach really is foolish. A warning supported by relevant evidence can save years of rediscovery. A failed attempt may be enough to stop.

What matters is preserving the distinction between receiving that evidence and pretending it was already possessed.

Every demand that someone should have known raises a question about how knowing was supposed to become possible. A sound demand points to responsibilities and conditions that improve the answer. An unsound demand enforces the appearance of having completed a learning process whose unfinished stages it makes harder to undertake.

Related:

• The Implicit Treaty — practical advice can name a solved state while omitting the knowledge and capabilities needed to reach it
• The Copenhagen Trap — unequal attribution of the costs of acting and not acting changes who participates
• Sterile Generativity — preserving recognizable outputs can undermine the capacity to produce valuable successors
• Information Is Not an Update — a message can arrive and leave the old model unchanged; the update is what the old model can no longer make disappear