Fixation in Complex Failures: What Signals Break a Locked-In Diagnosis?

Détail d'un panneau d'alarmes aux cellules ambrées allumées parmi des cases éteintes, cadre gris brossé

An alert appears during cruise, the crew recognizes it, applies the associated procedure, and the screen does not change. One minute later, two indications contradict the working hypothesis: they are seen, commented on, and then explained. No one has stopped looking at the screens: it is the reasoning that has closed. Three questions are worth asking: why the lock sets in, which signals reveal it in oneself and in others, and which triggers, already written into the procedures, break it.

The lock is in the hypothesis, not in the gaze

The term fixation evokes an immobile gaze on an instrument; this is almost never what happens. Under load, no one compares hypotheses in parallel: one is adopted, very soon, and is tested. This shortcut is not a weakness of character, it is what makes the work sustainable: without a working hypothesis, each parameter would have to be weighed separately when the load is already high.

The price is paid later. The hypothesis becomes the filter through which everything else passes: the indications compatible with it count as proof, the contrary ones receive an explanation. A suspicious sensor, a secondary consequence, a delay on the screen: each explanation costs a few seconds and preserves a model that was itself costly to build. Revising it, on the other hand, forces a complete reconstruction under time constraints. This asymmetry in cost is sufficient to explain the lock: no stubbornness is needed to find oneself in that situation.

Added to this is a learning effect: a well-trained crew quickly recognizes failures seen in training, and also recognizes them where they are not. Competence accelerates identification, and therefore commitment to one line of inquiry. That is why the exhortation to “stay vigilant” has no effect here: vigilance is intact, it is the interpretive framework that is frozen.

Once a crew picks a hypothesis, they see only what fits it. New clues are ignored or explained away, even if they don’t match. The brain does this to save time under stress, but it can lead to dangerous mistakes.

An A320 crew sees an engine vibration alert. They assume a sensor fault and follow the procedure. Later, fuel flow drops, but they explain it as a secondary effect. Only when the engine flames out do they realize it was a real failure—not a sensor issue. The hypothesis blinded them to key clues.

Overlap between the symptoms predicted by the hypothesis and those actually observedTwo ellipses overlap extensively. On the left, what the retained hypothesis predicts; on the right, what the crew observes. The common area is sufficient to sustain the hypothesis. The uncovered right part contains the contraindications, which receive an explanation instead of being integrated into the reasoning.what the hypothesis predictswhat is observednever foundconfirmscontraindicationswe discard what exceeds the model
A false hypothesis survives because it is partially true. The central overlapping area provides enough confirmations for the reasoning to close; the right area, unexplained, is the one that contains the decisive information.

Why the complex failure creates the lock

A simple failure is self-diagnostic: the cause, the symptom, and the remedy coincide, and the execution of the procedure closes the case. A complex failure behaves in the opposite way. Its signature is shared with other causes, manifests through a cascade of messages whose order depends on the alert logic and not on the chronology of the failure, and evolves: the situation read in the first minute is no longer the same in the third.

The alert system changes nothing, because that is not its role: it prioritizes, inhibits what is not relevant in the phase, directs towards a procedure, but does not designate a cause. The procedure, for its part, more often treats a consequence: isolates, reconfigures, protects. It can be executed completely and correctly while leaving the cause intact.

This is where the most costly confusion is born: the verified elements produce a trail of progress, something is advancing therefore something is being resolved, and the feeling of completion replaces understanding. When a new alarm arrives, it is read within the already established framework, not as a refutation, but as an expected consequence of the failure that is believed to be present.

The trap: a procedure that ends without anomaly does not prove that the diagnosis was good. It only proves that the requested actions could be executed. As long as the behavior of the aircraft has not changed in the direction expected by the hypothesis, the hypothesis is not validated: it simply remains standing.

Cost of review and value given to a contrary index, according to the time invested in a hypothesisTwo curves cross. In solid line, the cost of abandoning the hypothesis increases with the time invested. In dashed line, the value given to an index that contradicts it decreases. Beyond the crossing, the crew no longer spontaneously reviews.value of a contrary indexcost of reviewhighlowpoint of no returntime
The review window closes on its own. The more the crew has invested in a line of reasoning, the less weight a contrary clue carries: beyond the crossover point, the way out will no longer come from the reasoning itself, but from an external event.

Three signals that a diagnosis has closed

The lock is not announced with a visible error: it is announced with a change of style. Three markers are observable in real time, even in oneself, and do not require an instrument.

  • Repetition. The same action is taken a second, then a third time: a switch, a computer, a mode reset. Repetition is not a verification, it is the sign that no conclusion has been drawn from the first attempt and that the action is standing in for reasoning.
  • Closing language. The conditional disappears, “it would seem” becomes “it is”, and the remaining questions are rhetorical: they ask for confirmation, not information. A cockpit in which no one asks open questions is a cockpit in which no one expects surprises.
  • The gap explained away on the fly. Each contrary indication receives an immediate and plausible explanation: that is what makes it dangerous. The signal is not the falsity of the explanation, it is its speed: no one has taken the time to consider that the indication could be accurate.

A fourth, more discreet signal is worth listening to: time stops being called out: no one mentions fuel, remaining distance, the descent deadline. When temporal references leave the conversation, all available capacity has been absorbed by the diagnosis.

You don’t get out of a lock by will

Doubt cannot be ordered. Asking a crew to “question their hypothesis” is like asking them to spontaneously write off an investment they have just made. What works is of another nature: an event external to the reasoning, planned in advance, that does not need doubt to trigger.

The time limit is the most robust of these events, provided it is established before entering the diagnosis and announced out loud. The structured decision models used in operations do not all treat this resource in the same way: some place it explicitly at the beginning, which is the meaning of the T in T-DODAR, while others, built on establishing the facts before the options, leave it to the crew to set their own duration. The distinction is not cosmetic: a duration announced before the reasoning creates an appointment that does not belong to the hypothesis, and therefore one that cannot be absorbed by it. A duration decided afterwards, on the other hand, will be renegotiated: “thirty more seconds”, for as long as the hypothesis holds.

Other limits already exist in flight: leveling, entry into the holding pattern, fuel check, call from the senior cabin crew member, resumption of contact with ATC. Each one forces a reformulation for someone else, and it is the reformulation that reopens the reasoning.

The correct reflex: when the announced deadline arrives, do not ask “where are we?”: the question invites a summary, therefore a repetition of the hypothesis. Ask “what do we see that doesn’t fit?”. This formulation forces the production of a contrary element and makes it visible that there is none, which is information in itself.

Moment of hypothesis review with and without announced time limitTwo parallel timelines carry the same contraindications, represented by dots. Without an announced limit, the review only occurs at the end. With a limit announced beforehand, the reopening of the reasoning occurs early and frees up the time represented between the two moments.without time limitlate reviewcontraindicationswith announced limitlimitreopeningtime gained
It is not the accumulation of clues that triggers the review, it is the deadline. The same contrary indications appear in both cases; only the limit announced beforehand transforms their accumulation into a decision.

What really protects the division of tasks

The separation between the pilot flying and the pilot monitoring is not an organizational convenience, it is a barrier against fixation. It ensures that at least one person in the cockpit is not committed to the explanatory model being built, and therefore still has the ability to see a discrepancy. The day when both heads enter the same reasoning, this barrier disappears, and nothing signals it, since the cockpit seems very busy and very cooperative.

Two pilots in a cockpit at night, focused on the instrument panel and controls.
Focused attention prevents diagnostic oversights in high-stakes environments.

The merged diagnosis has two effects, and the second is the most serious: it eliminates the external witness, therefore any internal challenge, and leaves the flight path without continuous monitoring: altitude, speed, heading and energy are maintained only by automation whose authority the failure may precisely have reduced.

Two ways to manage a complex failure and what degrades first in each
What is observed Segregated diagnosis Merged diagnosis
Trajectory management One pilot designated, continuously No one continuously
Procedure reading By the pilot monitoring, out loud By both, over the shoulder
State of the hypothesis Declared by one, verified by the other Shared, therefore never questioned
Detection of a contrary indication Possible: one of the two is not committed Unlikely: both are
Exit from fixation Internal, through mutual control External only: cabin, ATC, alarm
What degrades first Available time Altitude, speed, position
Pilot workload during a complex failure, in segregated then merged modeFour bars represent the same flight segment. At the top, one pilot remains assigned to the trajectory while the other executes the procedure. At the bottom, both switch to diagnosis; the area framed in dashed line marks the period during which no one supervises the trajectory.segregated diagnosisPFtrajectory and communicationsPMprocedure and diagnosismerged diagnosisPFtrajectorydiagnosisPMdiagnosisno one is flying the aircraft
The change is not visible in the cockpit environment, it is visible in the schedule. As soon as both bars carry the same task, the flight path is maintained only by automation: the very systems that the failure may have degraded.

The external actor is underutilized. Cabin crew, ATC, remote maintenance support: their value does not lie in their expertise in the failed system, but in the fact that they have not invested in the hypothesis. Summarizing the situation for someone who has not seen anything forces the separation of the facts from the interpretation: that is what produces the effect, not the answer obtained.

The stimuli that really reopen the reasoning

Not all stimuli are equal: those that ask for a status report only get a repetition of the hypothesis, formulated with more certainty than the first time. Four formulations work better, because they force the production of something that the hypothesis does not contain.

  1. Declare the hypothesis with its degree of confidence. “It’s probably this, half sure” opens a space that “it’s this” closes. The degree of confidence declared out loud is an explicit invitation to contradict, and allows the other to speak without questioning the person.
  2. Ask for the contrary indication, not the confirmation. The question must demand an element that does not fit. Its repeated absence is in itself a result: either the hypothesis is solid, or no one is looking.
  3. Reason in terms of needs rather than failure. “What do we need to land this aircraft?” shifts the reasoning from diagnosis to management. Many failures are managed without being understood, and the question refocuses the crew without forcing anyone to publicly abandon their line of reasoning.
  4. Reread out loud the entry conditions of the current procedure. Mechanical stimulus, and one of the most effective: confronts the text with the screen and reveals cases where a procedure has been entered whose conditions are no longer met, or never were.
Diagnosis loop closed on itself and entry point of an external stimulusThree linked blocks—indices, hypothesis, action—form a cycle whose return only returns elements compatible with the hypothesis. Contraindications exit the loop and are discarded. An external stimulus, in dashed line, enters directly into the hypothesis block and reopens the cycle.indiceshypothesisactionconfirmation filterdiscardedexternal stimulusquestion that reopens
A closed loop does not reopen from the inside. The feedback from the action only returns what confirms; only an external entry point—deadline, third party, rereading of conditions—reaches the hypothesis block.

What this changes in training and checking

An evidence-based training program evaluates observable skills, not discoveries. However, many sessions still reward the crew that identifies the failure and penalize the one that does not identify it but protects its trajectory, manages its time, and preserves its options: the opposite of what the line demands.

The useful scenario, therefore, is not the rare failure, but the failure with an ambiguous signature: two possible causes, a procedure that treats a common symptom, a slow evolution. The session is judged by the existence of a turning point: did it happen?, what caused it?, and if not, what would have been needed? A debriefing that begins by revealing the failure kills the question: it is better to have the crew recount the order in which the clues appeared.

The symmetrical risk is real: training made solely of clean scenarios, where each failure has its answer, trains for fixation by instilling the idea that a found procedure is a problem solved. Organizations that take the subject seriously—you can find a panorama on the side of the FL330 training path—build their sessions on ambiguity rather than technical difficulty.

What to check in your own documentation

Three checks can be done on the ground, without an instructor, and pay off well. First, in the abnormal procedures that you apply most frequently, reread the entry conditions and the exit notes of the type “if such parameter does not return”. These are the only branches of the text that organize a reevaluation; knowing that they exist, and where, changes behavior in flight.

Then, identify which ones explicitly prescribe a time delay or a reevaluation, and which ones prescribe nothing: the latter are not less dangerous, they are silent, and it is up to the crew to establish their own limit. Finally, verify whether the operator’s decision model includes a formal review step, and whether it is actually practised in simulator sessions.

There remains a gray area that documentation rarely addresses: two conflicting abnormal procedures, or a procedure written for a single failure applied to a situation that includes several. No manual lists these combinations. What remains are the aviate, navigate, communicate hierarchy, the preservation of options—fuel, airports, configuration—and the obligation to say out loud that one is leaving the written framework, so that the decision is made by two and not imposed on one.

This article describes a mechanism; it does not create any procedure and does not replace anything. In case of discrepancy, the applicable regulatory text, the approved flight manual, and the operator’s documentation prevail over what is written here.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

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