Just culture: drawing the line between error, accepted risk-taking and negligence

Two aviation safety analysts review flight data in a dimly lit control room at dusk.

The report lands on a Tuesday morning: a take-off performed with a flap configuration that was not the one on the performance sheet, the discrepancy spotted in cruise by the crew themselves, who wrote it up and filed it. Nobody was hurt, the aircraft came back on schedule. The question that follows will occupy several people for weeks: an error, a risk the operator accepts, or a breach that calls for individual follow-up? Just culture is not a state of mind, it is a line of reasoning — and in that reasoning, the order of the questions decides the outcome as much as the facts do.

What is really being asked of the classification

A just culture policy promises neither impunity nor the courtroom. It promises a rule for dividing the ground: whatever stems from design, procedure, work organisation or training is not sanctioned, because sanctioning changes nothing there; whatever stems from a conscious choice to step outside a known limit is not let through, because letting it through moves that limit for everyone.

The stake is not moral, it is operational. The flow of occurrence reports is the only source of information on what actually happens on the line, and it is an asset that is quickly destroyed. Two files classified by gut feeling, two outcomes that seem to hinge on the person rather than on the act, and the volume of voluntary reports drops. It never climbs back at the rate at which it fell. The safety manager who loses that flow does not lose an indicator: he loses his sight of the operation.

Hence the requirement, carried by the European regulation on occurrence reporting and repeated in operators’ manuals: the line is defined in advance, it is written down, and it must produce the same result whatever the file and whoever handles it. A reproducible classification calls for three things: a fixed order of questions, stated tests, and announced timescales.

Four boxes, only one boundary that matters

Everyday vocabulary confuses the deviation with the fault. Four boxes are needed, not two.

Three men in uniform discuss a graph with intersecting lines and fictional text in a control room.
Balancing risk and accountability in high-stakes decision-making.
  • Error : the deviation was not perceived at the moment it occurred. A slip, capture by a habitual sequence, a mode error, a misreading of a displayed state. The individual could not decide not to commit it, since he could not see it.
  • Accepted risk-taking : the risk was perceived, weighed, and the deviation stays within the latitude the operator grants to that position. This is the heart of the job; an airline that allows no latitude does not obtain more safety, it obtains undeclared deviations.
  • Unaccepted risk-taking : the risk was perceived, and taken outside any foreseen latitude — often for a benefit that is not the operation’s: saving time, avoiding an explanation, sparing oneself a manoeuvre.
  • The intentional act : the harm itself is intended, or fitness was knowingly compromised. It falls outside the scope of just culture, which has never covered it.

The boundary that is costly to misplace is not the one separating error from deviation: it is the one separating accepted risk from unaccepted risk. The whole useful debate is played out there, and it is the only place where the file needs reasoning rather than a reading.

Continuum of behaviours, from error to the intentional actA horizontal band divided into four boxes from left to right: unintended error, accepted risk-taking, unaccepted risk-taking, intentional act. A rising dashed line above represents awareness of the risk at the moment of the act, low on the left and total on the right. A solid stroke marks the decisive boundary between the second and the third box. Below the band, two braces: the two boxes on the left call for a design or procedure fix, the two on the right open an individual question.awareness of the risk at the moment of the actdecisive boundaryerrorunintendedrisk-takingacceptedrisk-takingunacceptedactintentionalprocedural fixindividual questionthe choice stayed within the admitted latitudethe perceived risk was taken outside any latitudeerror is not sanctioned: it is made less likely
Only one of the three separations calls for reasoning. The boxes read from left to right according to awareness of the risk at the moment of the act. Moving from the second to the third changes the nature of the response; the other transitions merely refine the fix.

Just culture is about drawing a clear line between mistakes that happen to anyone and choices that put safety at risk. If someone makes a mistake they couldn’t see coming, the system needs to be fixed. If they knowingly take a risk that’s not allowed, they need to be held accountable. The tricky part is telling the difference between the two.

An A320 crew takes off with flaps set to 2 instead of 3, as shown on the performance chart. They notice in cruise and report it. The aircraft landed safely. The investigation finds that the crew misread the flap setting indicator in low light. This is classified as an error because a peer in the same situation could have done the same. The fix would be better lighting or clearer markings on the indicator.

The four behavioural regimes and the proportionate response each of them calls for
Classification Signature in the account Question that reveals it Proportionate response
Error The deviation was not perceived at the moment it occurred Could a peer placed at the same instant, with the same information, have done the same? Design, procedure or training fix; no individual follow-up
Accepted risk-taking The risk was perceived and weighed, the deviation stays within a recognised latitude Was that latitude foreseen, written down and known to the position? No follow-up; if needed, clarification of the latitude in the documentation
Unaccepted risk-taking Risk perceived, taken outside latitude, for the individual’s benefit more than the operation’s What did he gain by doing it that way, and did he know it was prohibited? Recorded interview, targeted remediation, temporary restriction according to severity
Deliberate or repeated breach The same deviation already notified to this individual, remediation already carried out and recorded, repetition What has happened since the previous time? Individual follow-up owned and justified in writing

The classification tree, and why its order is locked

The reasoning comes down to four questions asked in an order that is not negotiable. Every positive answer stops the descent and closes the next debate.

First: was the harm intended, or was fitness knowingly compromised? At that point one leaves the framework. Only then: could a professional of the same body, of the same qualification and comparable experience, placed in the same situation, have acted in the same way? If yes, the act is an error and the fix is collective. If no: did the individual have the training, the procedure, the tool and the time to act otherwise? If one of these is missing, we are still facing a system failing. And last of all, only then: had this deviation already been notified to this individual, and had remediation been carried out and recorded?

Starting with the last question — the prior record, the reputation, the “we know him” — amounts to judging the person before the act. It is the most frequent investigative fault, and the most destructive for the flow of reports, because it is visible to everyone who works with the person concerned.

Classification tree in four questions asked in orderFour questions stacked on the left, each linked to the right by a “yes” arrow towards an outcome: intended harm leads to the intentional act, outside the framework; would a peer have done the same leads to error and a collective fix; means of acting otherwise leads to a system failing; deviation already notified and remedied leads to repeated breach. Each “no” steps down one level; at the bottom, the default outcome is risk-taking to be examined in its context.harm intended?would a peer have done the same?means of acting otherwise?already notified and remedied?risk-taking to be examined in its contextintentional act, outside the frameworkerror: collective fixsystem failingrepeated breachyesyesyesyesnononono
Every “yes” stops the descent. The individual’s prior record only comes in last, once the act has been classified on its own terms. Moving that question to the start turns the investigation into a judgement of the person.

The substitution test, and the three ways of distorting it

The substitution test is the central tool, and the most mistreated. Its exact wording matters: you replace the person, you freeze everything else. Same qualification, same experience, same crew, same machine, same information available at the same instant, same time pressure, same fatigue, same weather, same procedure in force that day. The question is not “what should a good professional have done?” but “what could a comparable professional have done in that particular situation?”.

Three manipulations, almost always unintentional, make it unusable. The first consists in giving the peer being questioned what the investigation produced: the flight parameters, the transcript, the outcome. No crew has that in real time. The second consists in choosing an unrepresentative peer — an instructor in the field concerned, a manager, the in-house expert — whose answer says nothing about what the line actually does. The third consists in asking the question in a quiet room, pen in hand, without any of the constraints that made it difficult.

When the test answers “no, a peer would not have done that”, it does not yet conclude. It opens the complementary test, that of the means: training received, procedure available and applicable, tool in working order, physical time to act otherwise. As long as one of these elements is missing, responsibility remains organisational.

What the substitution test freezes, and how its answer readsAt the top, four boxes represent the elements to be kept identical: information available, pressure and workload, procedures and training, state of the equipment. An arrow points down to a horizontal axis graduated from none to all, measuring the share of comparable peers who would have acted the same way. The left-hand zone is labelled individual deviation, the central zone grey area to be examined, the right-hand zone system failing to be dealt with at procedure level.you replace the person, you freeze everything elseinformationavailablepressureand workloadproceduresand trainingstate ofthe equipmentindividual deviationgrey areasystem failingfew peersdivided answersmost peersnoneallshare of comparable peers who could have acted the same waythe test does not say whether the act was right: it says where to apply the fix
The test result designates a level of action, not a culprit. The higher the proportion of peers who would have acted the same way, the further the fix must move up towards the procedure and the organisation. A massively affirmative answer is a signal of drift, not an exoneration.

The right reflex: reconstruct in writing, before any interview, the chronology of what was available at the moment of the decision — what was displayed, what had been passed on, what could not be known — then submit that chronology alone to the peer being questioned. Everything learned after the event stays off the sheet.

Two hindsight biases that are systematically confused

The first is hindsight bias: once the outcome is known, the path leading to it looks obvious and the other branches become invisible. The crew, for their part, faced a tree: several plausible hypotheses, no certainty, a constrained decision time. The person handling the file, meanwhile, looks down a corridor. He cannot give himself back the uncertainty, he can only remember that it existed, and write it down.

The second is outcome-severity bias, more insidious because it cannot be felt. With strictly identical conduct, the more serious the outcome, the harsher the judgement passed. Two reports describing the same unstabilised approach, one ending in a go-around, the other in a runway excursion, will rarely come out of the investigation with the same classification — even though the conduct to be classified is the same. There is a simple check: rewrite the file changing only the outcome, and redo the classification. If it changes, it is not the act that is being judged.

Harshness of judgement as a function of the severity of the outcome, with conduct held constantA chart with the severity of the outcome on the horizontal axis, from no consequence to accident, and the harshness of the judgement passed on the vertical axis. A solid curve rises steeply from left to right: this is the judgement passed after the fact. A horizontal dashed line stays flat at mid-height: this is the conduct actually observed, which does not vary. The gap between the two traces widens with severity.harshness of judgementseverity of the outcomejudgement passed after the factconduct actually observed, unchangedno consequenceincidentaccident
The rising curve is not the curve of the act. The conduct to be classified is the same across the whole width of the chart; only the outcome changes. The gap between the solid line and the dashed line measures exactly what outcome-severity bias adds to the file.

The trap: the investigation produces information the crew did not have — recorded parameters, communications, the real state of a piece of equipment — and that information silently contaminates the classification. One ends up faulting them for not knowing what only the recording reveals. Any item derived from the exploitation of recordings must be marked as such in the file, and kept out of the substitution test.

When risk-taking has become the local norm

A deviation that works out does not remain a deviation: it becomes a reference. Next time, the same shortcut starts from a threshold that has already moved, and nothing in the system flags that shift since, by definition, it produces no occurrence. Practice drifts away from the written procedure without any decision ever having been taken, and the deviation only shows up at the moment the remaining margin is exhausted — that is, at the moment of the occurrence.

Classifying individually the person who happened to be there that day is both unfair and ineffective: unfair because he was applying the practice of his own sector, ineffective because the practice will remain after him. This is precisely the case where the substitution test answers “most would have done the same”: that answer exonerates nobody, it changes the addressee of the fix. It points to an unworkable procedure, a missing check, or a production objective that nobody has ever set against the procedure.

Negligence, the real thing, has a different and far narrower signature: the same deviation, already notified to this person, with remediation carried out, recorded and accepted — then started again. There, the individual response is not merely legitimate, it is expected by colleagues, who are perfectly able to tell the difference.

Drift of practice away from the written procedure, up to the occurrenceA chart with time on the horizontal axis and the gap from procedure on the vertical axis. A low dashed line represents the written procedure, which does not move. A second, higher dashed line represents the threshold beyond which there is no margin left. Between the two, tolerated practice rises in successive steps, each step corresponding to a deviation that worked out, until it touches the upper threshold where a dot marks the occurrence.gap from proceduretimethreshold beyond which there is no margin leftwritten procedure, unchangedtolerated practiceoccurrence
Every step is a deviation that ended well. The procedure does not move, practice rises, and the system sees nothing as long as the margin absorbs it. The individual present at the last step invented nothing: he inherited the position of the cursor.

Handling a file without drying up the flow of reports

The classification is only worth as much as the framework around it. Four conditions determine it, and they can be checked on documents.

Separation of roles. The person who analyses the occurrence for safety is not the one who decides on individual follow-up. Safety management system data do not travel up the management line; they feed the analysis, and nothing else. When that watertightness does not exist, word gets round very quickly across the whole body concerned, and the content of the reports thins out even before their number falls.

Timescale and feedback. An open file must have an announced closing date and a reasoned decision, communicated to the reporter — including, and above all, when the decision is that nothing happens. Silence is never read as an absence of follow-up: it is read as a threat left pending.

The choice between anonymity and confidentiality. This is not a detail on a form: it is what determines whether the substitution test will be workable.

Anonymous report and confidential report: two arrangements that do not produce the same result
Criterion Anonymous report Confidential report
Calling the reporter back Impossible: the analysis stops at the text received Possible through a single point of entry, sole holder of the identity
Substitution test Rarely workable, the context is missing Workable, the context is reconstructed with the person concerned
Trust produced Immediate but fragile: it rests on the absence of any record Slower to establish, more solid: it rests on a written and enforceable rule
Appropriate use Sensitive subjects, degraded climate, starting up an arrangement Standard regime of an established safety management system

The panel and its composition. A classification that leads to individual follow-up is decided by a mixed panel, with representation of the body concerned — a cabin crew file that contains the opinion of no cabin crew member will be challenged on that ground alone, and rightly so. These subjects are usefully discussed between peers from different operators, for example on the professional exchanges of the sector, where compared policies are worth more than general principles.

At a small operator, these four conditions run into a practical obstacle: the safety manager is often also the chief pilot or the nominated person for flight operations, and separation of roles cannot be obtained from the organisation chart. It is then built through procedure — the file handled with a peer from outside the organisation, named in the manual and mandated in advance, or a cross-opinion from a partner operator on the files that lead to individual follow-up alone. A training organisation, an aerial work operator or a business aviation structure do not need a lighter arrangement than an airline’s: they need a third party. Without one, classification and sanction come out of the same mouth, and the flow dries up all the faster because everyone knows everyone.

What you should be able to pull out of your own documentation

The quickest test of the maturity of an arrangement comes down to five questions put to the operator’s documentation, without going through anyone. Is there a just culture policy document, signed both by management and by the staff representatives, and published? Does that document name who classifies, within what timescale, and following what sequence of questions? Is the use of flight data covered by a separate agreement, with a de-identification procedure and a designated point of contact? Does the manual set out what happens when an authority or a court asks for an item held in the safety management system? And is there an appeal for the individual who contests a classification?

One area remains grey, and it must be named rather than promised as settled: the protection organised by the operator and by the European regulation on occurrence reporting does not close off the judicial route. One and the same occurrence can follow two channels, one of which escapes the airline entirely, with its own classification criteria, which are not those of just culture. The role of the internal arrangement is not to make that risk disappear: it is to add nothing to it, and never itself to become the source of the evidence.

For anyone preparing to take up a safety post or facing a selection interview, what is expected is not the recitation of a definition. It is being able to run through the tree in three sentences, to explain why the question of prior record comes last, and to cite a case where the substitution test moved a file from the person to the procedure. That is the capability which distinguishes a safety manager from a writer of reports.

The applicable text and the operator’s approved documentation prevail over this article: in case of divergence, they are what counts, and the just culture policy signed within the company alone governs how a file actually unfolds.

Questions fréquentes

Does the substitution test also apply to maintenance and cabin crew personnel?

Yes, provided the person is replaced by a genuinely comparable peer: same certification, same experience, same team, same working conditions at the moment of the action. For a maintenance task, that includes the state of the documentation used, the lighting, the working position and the point in the shift. A test run with a more experienced technician or in ideal workshop conditions proves nothing.

Can a deviation committed for the first time justify individual follow-up?

Yes, but only in two cases: the intentional act, and conscious risk-taking outside any foreseen latitude, where the individual clearly had the means to act otherwise. Outside those cases, a first occurrence of a deviation calls for recorded remediation, not a sanction. Repetition after remediation changes the classification; the first time does not.

What should be done when the substitution test answers that the majority of colleagues would have acted the same way?

That result closes the individual question and opens a procedural one. It indicates that the actual practice of the sector has drifted away from the written text without anyone having decided it, which is the classic sign of normalised drift. The treatment then bears on the procedure, its feasibility, the associated workload and the checks that should have made the deviation visible sooner.

Does a just culture policy protect against judicial proceedings?

No. It organises the internal use of safety information and limits what the operator may do with it, but it does not close off the judicial route, which follows its own criteria. The useful protection consists in the internal arrangement never itself becoming the source of the evidence, and in the manual explicitly setting out what to do in the event of an outside request.

Should anonymous reporting be preferred in order to obtain more reports?

Anonymity increases volume in the short term but severely degrades usability: with no way of calling the reporter back, the context is missing and the substitution test becomes impracticable. The standard regime of an established system is protected confidentiality, with a single point of entry holding the identity. Anonymity keeps its value for sensitive subjects or for starting up an arrangement in a degraded climate.

How can outcome-severity bias be neutralised in an investigation?

The most effective check consists in rewriting the file changing only the outcome, then redoing the classification. If the conclusion moves while the conduct described is identical, it is the bias that spoke and not the analysis. Marking in the file any information derived from the recordings, unavailable to the crew at the time, produces the same corrective effect.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

FL330, c’est toute l’aviation professionnelle — explorez nos autres espaces.