A servo actuator removed to replace seals, reinstalled at the end of the shift, and a missing safety wire discovered two days later at the transit check. The internal report reads in one line: ‘assembly error, mechanic made aware.’ That line closes the file and protects no one, because it names the last link as if it were the first. Here is the method for tracing back from the reinstalled part to the local conditions, and then to the process that produced them.
The act is an entry point, never a conclusion
The chain followed by the shop is logical: safety wire missing, therefore an omission on reassembly, therefore an operator to be retrained. It is wrong — not because the omission did not occur, it is documented and acknowledged, but because it stops at the only part of the event that will never recur in exactly the same form.
An action aimed at the act addresses one person, at one moment. It is immediate, visible in a meeting, and it evaporates: that person will change teams, the position will be held by someone else, and the condition that made the omission likely will still be there. An action aimed at local conditions — access to the workstation, the availability of the correct-diameter lockwire, the task-interruption rule — addresses everyone who will work there. An action aimed at the process — the breakdown of the work card, shift planning, the handover rule between teams — addresses every task in the same family, at every base, and people who have not yet been hired.
When something goes wrong, blaming one person doesn’t fix the real problem. The issue might be where they work (like bad lighting or missing tools) or how the work is organized (like unclear instructions or too many interruptions). Fixing those things helps everyone, not just that one person.
On an Airbus A320, a mechanic missed a safety wire during a servo actuator repair. The issue wasn’t just the mechanic’s mistake – it was because the work card didn’t have a separate signature box for safety wiring, and the mechanic was interrupted near the end of their shift. Fixing the work card and adding a mandatory break before shift changes would prevent similar mistakes in the future.
The first sort of a report therefore does not consist of assigning blame, but of placing each finding in one of these three strata. As long as the file contains only findings from the first, the analysis has not begun.
Reconstructing the task before judging the act
Before any hypothesis, the task is written up as it was actually carried out, not as it should have been. This means quickly recovering perishable elements: the work card in its original state — not a clean reprint —, the timestamps of the signatures, the log of tooling issued and returned, the actual shift assignments, the trace of parts requests. After a few days, some of these elements will have been filed away, overwritten, or reconstructed from memory — and reconstruction from memory is already an interpretation.
In the case at hand, the reconstruction gives the following. The removal and replacement of the seals proceed without incident. Reassembly begins as the shift is drawing to a close. A call diverts the mechanic to another task judged more urgent; he is away for about twenty minutes, returns, picks his card back up, and resumes work at the point he believes is his: the next step.
This last point is not a personal weakness but an effect well described by research on task interruption. On resumption, an operator does not restart from where he actually stopped, he restarts from where he believes he stopped, and the bias runs forward: what is retained in memory of an interrupted task is not the step in progress, it is the intention for the next one. The safety-wiring was logged as done because it had been planned.
Local conditions are not excuses
The established facts are then listed and read twice: once as a judgment on the act, once as a question put to the system. The first reading produces a sanction, the second produces an action. The table below applies the exercise to the case.

| Established fact | Reading that stops at the act | Question that opens the process |
|---|---|---|
| The safety wire was not fitted | Omission on reassembly | Is safety-wiring a distinct step on the card, with its own signature box? |
| A call diverted the mechanic mid-task | Lack of rigor, he should have finished | Who has the authority to interrupt a task in progress, and is that rule written down anywhere? |
| Resumption happened without rereading the card | Individual negligence | Does the card require a formal resumption point after an interruption? |
| The task was finishing at the end of the shift | Fatigue | Does the scheduling provide for a mandatory task end or a logged handover? |
| The inspection signed off the whole job | Double human error | Does the independent inspection cover the critical step or the closing of the file? |
| The workstation was lit by a work lamp | Degraded conditions accepted | Are the access and lighting for this workstation specified in the documentation, or improvised each time? |
The trap. Fatigue is the contributing factor easiest to write down and the hardest to address: it explains everything, so it explains nothing. Until it has been linked to an identifiable decision — a long task started with no margin at the end of a shift, a team handover not provided for by the procedure — it is not a contributing factor that has been produced, it is a comment. The test is mechanical: a contributing factor must be capable of turning into a change to a document, a tool, or a rule. Otherwise it has no place in the report.
Interruptions cannot be eliminated: they are the normal regime of a base in operation, and a procedure banning them would be circumvented on day one. What can be addressed is their unit cost. An interruption on a task where each step carries its own signature box does not produce the same effect as an interruption in the middle of a continuous paragraph signed only once at the end.
The right reflex. When you are interrupted, do not leave the workstation before physically marking the card at the current line, and leave the tool or part visibly positioned in a way that does not look finished. On resumption, restart from the previous step, never from the one you believe comes next — rereading a step already done costs a few seconds, discovering a skipped step late costs a flight.
The work card is evidence
A card that groups ‘reinstall, safety-wire, inspect’ under a single signature box offers no stopping point at the exact spot where the risk is concentrated. How a card is broken down is not a matter of form or drafting convenience: it determines where attention is forced to stop, and therefore where an interruption can be caught.
The card is therefore requested in its exact state at the time of the events, with its revision index, and compared against the approved maintenance data it derives from. Three discrepancies recur: the critical step buried in continuous text when the source data made it a distinct operation; the local rewording that lost a value, a limit, or a condition of application; the outdated revision left in circulation because the update stopped at the engineering office without reaching the workstation.
The same event does not read the same way depending on where the work took place, either. In the hangar, the interruption most often comes from the organization itself — a reprioritized job, an awaited part, a shift coming to an end — and is first addressed through planning. On the ramp, it comes from outside: a departure to be kept, fueling in progress, a crew waiting. The room for maneuver over the causes is small there, which shifts the effort toward the robustness of the card and the resumption rule, the only variables that remain under the organization’s control. An action plan copied from the hangar to the ramp fails for this reason — and so does the reverse: in the hangar, settling for a formal resumption point alone leaves in place the scheduling that manufactures the interruptions.
None of these three discrepancies appears in a report that starts with ‘assembly error.’ All three become obvious as soon as the real card is placed next to the source data.
From the isolated case to the aggregate
Not every contributing factor deserves the same response, and an organization that treated each one with the same energy would wear itself out without treating anything thoroughly. Two axes are enough to sort them: is the condition common or exceptional, and would the error it produces be caught downstream by an existing barrier?
An isolated report is handled on its own; it is the aggregate that informs. But the classification must be stable: if the same condition — unmanaged interruption, illegible workstation documentation, unavailable tooling — is coded three different ways depending on the base or the analyst, the aggregate will never show anything. A short, mandatory taxonomy is worth more than a rich one that no one applies the same way.
A second point of vigilance: the feedback loop. A system whose reporter never hears back again loses its throughput, and the loss is silent — the number of events drops, which management readily reads as an improvement. The average delay between filing a report and feedback to the reporter is a far more reliable health indicator than the volume received.
Just culture: the boundary is tested, not decreed
No one reports an error whose handling they cannot predict. Just culture is not the absence of consequence — an organization that never sanctions anything loses its credibility just as surely as one that sanctions everything: it is a written boundary, known in advance, applied the same way regardless of the outcome.
Three principles make it workable. The order of the questions, first: intent is addressed before everything else and, once ruled out, does not come back into the discussion. Indifference to the consequence, next: the same act, whether caught at inspection or discovered after a flight, receives the same qualification, otherwise the system quietly teaches that it is better not to report what went wrong. The separation of people, finally: whoever analyzes the event does not decide any individual follow-up, and the organization must be able to show this, not merely assert it.
There remains repetition. An isolated deviation falls under conditions; the same deviation, by the same person, once the conditions have been corrected, falls under something else. The practical difficulty is distinguishing the repetition of an individual from that of a position: in a hangar, the latter is far more frequent, and it is regularly read as the former because it is the person’s name that appears on the card.
What to go and check in your own documentation
Five points can be checked without a formal audit, from documents already in circulation, and they say most of what matters about an organization’s maturity. Does the task-interruption rule exist in writing, and does it specify who may interrupt whom? Do the cards carry a signature box at the safety-wiring, torque, or locking step, or only at the end of the block? Is the independent inspection defined by the operation to be verified or by the file to be closed? Is the taxonomy of contributing factors common to all bases, subcontractors included? Is the delay before feedback to the reporter measured, and by whom?
Two gray areas remain, and it is better to name them than to discover them in an audit. First, the task shared between two shops: the handover of an unfinished task is the moment when the two reporting systems stop talking to each other, and it is almost always at this seam that the analyses stop. Second, the status of temporary or seconded staff, whose reports are protected by the host organization’s system, but whose contract depends on a third party that has signed no just-culture charter.
For anyone preparing to take on a role in quality or in event-report management, this is where the difference lies between knowing the text and knowing how to make it live: the text describes a system, the exercise consists of showing where it leaks in your own organization. The FL330 training courses work through this mechanism using cases of this kind.
This article describes a method of analysis, not an applicable requirement. The regulation in force, the approved maintenance data, and your organization’s approved documentation prevail over anything written here, whether in case of divergence or of silence.
Questions fréquentes
What is the difference between a contributing factor and an excuse?
A contributing factor turns into a change to a document, a tool, or a rule. If it leads to no possible action, it is a comment. Fatigue alone, for example, only becomes a contributing factor once it is linked to an identifiable planning decision, such as a long task started with no margin at the end of a shift.
Should task interruptions be banned in the hangar?
No, such a ban would be circumvented from day one, because interruption is the normal regime of a base in operation. What can be addressed is the cost of an interruption: a card broken into individually signed steps and a resumption rule requiring a restart from the previous step make the gap detectable. The goal is to make interruption harmless, not rare.
How can the substitution test be applied without turning it into systematic leniency?
The test compares the operator to a peer with the same training and the same experience, placed in the same conditions — not to an ideal operator or to the analyst who already knows the outcome. It is applied only after intent has been ruled out. If it concludes that a peer would not have done the same, the case shifts toward aptitude, training, or supervision, which is not an absence of consequence.
Should the severity of the consequences weigh in how a deviation is qualified?
No. The same act, whether caught at inspection or discovered after a flight, must receive the same qualification. Letting the response vary with the consequence quietly teaches that it is better not to report what went wrong, and dries up the reporting system.
How can you tell whether the internal reporting system is still working?
The volume of reports received is a poor indicator, because a drop is easily read as an improvement when it often signals disengagement instead. Measure instead the delay between a report being filed and feedback being given to the reporter, along with the consistency of contributing-factor coding from one base to another. These two indicators predict the health of the system.
What should be checked first on a work card after an event of this kind?
Recover the card in its exact state at the time of the events, with its revision index, and compare it against the approved maintenance data it derives from. Three discrepancies recur: the critical step buried in continuous text, the local rewording that lost a value or a limit, and the outdated revision index left in circulation at the workstation.

Leave a Reply
You must be logged in to post a comment.