When Flight Data Contradicts the Crew’s Report

Formulaire manuscrit retourné à côté d'un ordinateur affichant des tracés de données abstraits

The report filed by the crew reads: « configuration exit compliant, nothing to report. » Flight data monitoring, for its part, recorded on that same flight a configuration-limit exceedance lasting a few seconds. Both documents are sincere, they do not tell the same story, and a decision must be made about what to do with that without damaging what allowed the report to exist in the first place.

A discrepancy is almost never a lie

Two observation systems coexist, and they do not observe the same object. The recorder samples parameters at a fixed rate: it knows the exact second a value was exceeded, but it knows nothing of what was displayed, what was said, or what had just been interrupted. The report, for its part, renders a lived experience: what the crew perceived, understood and decided, put into words several hours later, often at a hotel, from a memory that has already been reshaped. Expecting the two to coincide is like expecting a photograph and a testimony to overlap pixel for pixel.

Most exceedances flagged by a monitoring programme leave no perceptible trace on the flight deck. No alarm, no change in sound, no vibration: a parameter exceeded by a small margin, for a few seconds, leaves no conscious trace. The crew is not concealing the event — it did not experience it as an event. And even when a display does show the quantity in question, what it shows is not what is recorded: the value presented is filtered, damped, rounded to the unit useful for flying, whereas the recording retains the raw sampled value.

An additional effect is well known to anyone who collects accounts of an event: under heavy workload, the duration of a sequence is most often underestimated, and the order of events reorganises itself around what made sense. A sequence lasting several tens of seconds is reported as « immediate, » a reflex action is remembered as having preceded the alert that triggered it. This is not a failing of the individual — it is the normal functioning of a memory that encodes meanings, not timestamps.

The flight recorder and the crew’s report are two different ways of seeing the same flight. The recorder captures raw data, while the report captures human perception. They won’t match perfectly, and that’s normal.

On a Boeing 737, the flight data recorder might show a brief exceedance of the maximum landing gear extension speed (e.g., 280 knots instead of 250 knots) during a go-around. The crew, under high workload, might not notice this exceedance because no warning appeared on their displays, and they would report a normal procedure. The data shows the exceedance, but the crew’s report reflects their actual experience.

Reported duration of a demanding sequence versus its measured durationA chart whose horizontal axis carries the measured duration of a sequence and whose vertical axis carries the duration reported by the crew. A dashed diagonal represents a perfect match between the two. A solid curve starts at the same point, bends quickly and stays below the diagonal, the gap widening as the sequence lengthens: the longer the event lasts, the shorter it is reported as being.gapreported durationmeasured duration of the sequenceperfect matchwhat is reported
The gap widens with duration, not with bad faith. The diagonal represents a report that would match the measurement exactly; the curve represents what a memory formed under load actually returns. A long sequence is almost always reported as short, and the size of the gap tells you first about the load endured.

It is also worth remembering the initial gesture. Someone wrote the report, then filed it. In a system that works, that gesture is voluntary, it costs time at the end of a duty period, and it exposes the person who makes it. Treating it as a statement to be verified makes it costly a second time — and the second time, it no longer happens.

Three families of discrepancy, only one calls good faith into question

The first family stems from the measurement itself. Flight data is not truth: it is the product of a chain — sensor, bus, recorder, extraction, parameter dictionary, detection algorithm. A parameter that is computed rather than measured, a sampling rate that is low on one channel and high on another, an offset of around a second between two independently time-stamped sources are enough to manufacture an exceedance that never happened, or to conceal a real one. An algorithm calibrated to a nominal value when the aircraft was flying with a particular configuration or weight will produce the same illusion.

A man in a dark uniform monitors flight data on a cockpit screen.
Flight data can reveal discrepancies in crew reports.
Recorded value and displayed value during the same exceedanceA parameter plotted against time. A horizontal dashed line represents the limit. The solid curve, drawn from the recording, crosses this limit for a few seconds, and the resulting area is marked. The dashed curve, representing the damped value shown on the flight deck, follows the same shape but stays below the limit for the entire sequence: nothing the crew sees signals the event.recordedshown on the flight decklimita few secondsparameter valuetime
What is recorded and what is seen are not the same quantity. The indication shown on the flight deck is damped and rounded to be flyable; the recording retains the raw value. A brief exceedance can therefore be entirely real and entirely invisible to the pilot at the controls.

The second family, by far the most frequent, stems from perception and memory. The account is sincere and it is wrong on one point: a duration, an order, a value, the attribution of an action to one crew member rather than the other. Chronological inversion is the most confusing for a novice analyst, because it looks like a self-serving reconstruction when it is actually the mechanical result of encoding by meaning.

Same event, two orders: recorded chronology and recalled chronologyTwo horizontal timelines each carrying three points. On the upper timeline, the data timeline, the alert precedes the action, which precedes the return to normal. On the lower timeline, the timeline from the account collected, the action is reported before the alert: two crossed dashed lines connect the corresponding points and make the inversion visible, while the third point stays in place.dataalertactionreturn to normalaccountactionalertreturn to normaltime
The reported order follows the logic of the action, not the clock. The gesture that resolved the situation is remembered as the cause of the sequence, and so is placed first. An inversion of two events separated by a few seconds is a sign of workload, not a sign of concealment.

The third family is deliberate omission. It exists, it is rare, and it is not recognised by the size of the discrepancy. It is recognised by its shape: an account that answers an objection no one raised, that justifies before being questioned, or that contradicts an undisputable and easily verifiable fact — a position, a time of contact, a manoeuvre requested by control. An isolated discrepancy over a duration never carries this signature.

Three families of discrepancy between data and account: what distinguishes them and where to start
Family What is observed Clue that points the way First action Classic mistake
Measurement discrepancy The data asserts an exceedance that nothing else corroborates Computed parameter, low-rate channel, threshold set to a nominal value Replay the case through the extraction chain before any contact Opening an event file on a decoding artefact
Perception discrepancy Consistent account, but duration underestimated, order reversed, or action wrongly attributed Heavy workload, interruption, short phase, no indication on the flight deck Context interview, without showing the trace up front Treating imprecision as a contradiction
Intent discrepancy The account contradicts a verifiable external fact Unsolicited justification, inconsistency with a third-party source Exit routine processing per the written procedure, without improvising Jumping to this hypothesis at the first sign of divergence

Protection covers use, not accuracy

This is the point the practitioner must hold firmly. The voluntary report is protected because it was filed in good faith, not because it is accurate. The flight data monitoring programme is non-punitive by design: de-identification is its cornerstone, access to identity is reserved to a designated data holder, under written conditions usually agreed with flight crew representatives. None of this presupposes that the account has been verified; the architecture, on the contrary, presupposes that it will not be.

Protection has limits, which the texts state in restrictive terms: an intentional act, a manifest and gross breach of the rules, use of the information for purposes unrelated to safety — the wording that governs being that of the applicable text and the operator’s procedure, not that of a summary. A discrepancy between a trace and an account falls into none of these categories. It is not an established fact, it is an open question. Confusing it with a breach of the duty to report is the costliest mistake, because it is irreversible: a programme that has once checked an account against the recording receives from then on only accounts aligned with what the recording might show.

The trap: treating the discrepancy as a fault in the report. As soon as the crew understands that its text serves as an item to be checked against, the report becomes a defensive exercise: neutral sentences, no context, no mention of doubt or hesitation. The volume of reports is preserved and exactly what made them useful is lost.

Running the reconciliation without turning it into an interrogation

The question is not who is right, it is recovering what is missing from each of the two sources. This requires a precise conduct. Who makes contact: the data holder or an authorised peer analyst, never the line of command, and never a department whose purpose is disciplinary. When: early enough that the context is still accessible, late enough not to catch a crew at the end of a duty period. How: by announcing the framework up front — what is being sought, what will be written down, what stays anonymous, who will read the document, and what will not go any further.

One detail of conduct changes the entire yield of the interview: the trace is not shown first. As soon as it is placed on the table, the account aligns itself with it, the crew fills in the blanks with what the curve suggests, and independent information disappears. First the sequence is collected as it was lived, noting what was displayed, what had just been requested, what had just been interrupted — only then is the recording presented, and the crew asked what it brings to mind. Both versions then remain usable; in the reverse order, only one does.

The right reflex: ask context questions, never justification questions. « What was displayed at that moment? », « What had just been asked of you? », « What was on your mind at that point in the descent? » open the file. « Why didn’t you… » closes it, and closes the next ones too, because the question travels from one crew to another far faster than the analysis note.

Next steps according to the size of the discrepancy and what the interview yieldsA two-way matrix. In columns, the size of the discrepancy between the account and the data, small then large. In rows, the outcome of the interview: context explains the discrepancy, or it does not. The four cells give the next step, from a note filed in the record up to a review of the measurement chain, and, in the single case where nothing explains a large discrepancy, referral to a committee — the source remaining protected in every case.small discrepancylarge discrepancycontextexplains itnothingexplains itreconciliation notefiled in the recordcontributing factorsfed into the analysisreview of themeasurement chaincommittee reviewsource protected
Only one cell exits routine processing, and it does not exit alone. A large discrepancy that no context explains goes before the designated body, under the written procedure. The other three cells close within the programme, without ever lifting anonymity.

Which version feeds the analysis

Both, but not in the same part of the file. The data has authority over the what and the when: chronology, values, duration, configuration. The account has authority over the why: what was known, believed, displayed, said, the workload at the time, the interruption, the time pressure, the fatigue level, coordination on the flight deck. An analysis that discards the account on the grounds that it is inaccurate keeps a dated event and loses all the causal material; it then produces a hollow corrective action — a reminder of the limit in an operations circular — which everyone knows in advance will change no trajectory.

Before taking any action, the measurement itself still needs to be validated. Check that the parameter in question is measured rather than reconstructed, that its sampling rate supports the conclusion drawn, that the algorithm’s threshold does indeed match the limit applicable to that aircraft in that configuration, and that the channels being compared share the same time reference. This check is short compared with everything that follows, and it avoids convening a collective review around an extraction artefact.

The two sources and where they meet in the analysis fileTwo boxes at the top: on the left, the discrepancy detected in the flight data; on the right, the report filed by the crew. Two arrows converge on a central box, the reconciliation note, where identity remains protected. A third arrow descends from this box to a final box carrying the contributing factors and the actions adopted. Neither source replaces the other: they meet without correcting each other.discrepancy detectedin the flight datareport filedby the crewreconciliation noteidentity always protectedcontributing factorsand actions adopted
Two inputs, one note, no substitution. The data provides the chronology and the values, the account provides what was known and perceived. Reconciliation holds them side by side and records what remains unexplained, instead of ruling in favour of one over the other.

And the report is not corrected. The crew is not asked to rewrite it, no corrective note is added to it: this document is testimony of a perception at a given moment, and its value lies precisely in the fact that it was never revised. A dated reconciliation note is added instead, setting out what the data shows, what the interview brought, and what remains unexplained. This last heading is the most useful item in the file: it is what the operator knows it does not know.

What to go and check in your own documentation

On paper, every operator has the same safeguards; in practice, robustness comes down to four points that can be checked quickly, documentation in hand. Who is named as the data holder, and what happens during their leave. How long the elements that would allow tracing back to a crew are retained, and who carries out their deletion. Under what written procedure, and on whose decision, a file exits routine processing. And finally, how the data programme and the reporting system fit together: two separate files that only meet at the point of reconciliation, or a single file where one becomes the evidence against the other.

A grey area remains, and it is better to have settled it in advance: the flight with no usable data. A fleet not covered by the programme, a parameter absent from the recording set, a missed extraction, a recording overwritten before retrieval. The account is then the only source. The standard of proof cannot vary according to whether a recording is available, on pain of instituting two regimes within the same fleet — and crews spot this kind of asymmetry far faster than procedure writers imagine.

The arrangement also changes with the type of operation. In a small structure — aerial work, on-demand transport, a fleet of a few aircraft — the data holder, the analyst and the designated safety manager are often the same person: the separation of roles that protects the source no longer exists, whatever the manual says. It is restored through a third party, a peer from another operation or a mandated body, never by a simple written statement. In a charter or wet-lease arrangement, the difficulty runs the other way: the two sources live in two separate organisations, the data staying with the operator while the report can travel up to the contracting party. What is reconciled with what, and what crosses the boundary between the two systems, must then have been settled before the event — without that, the only item that circulates is the one that identifies the crew.

For anyone preparing to take up a post in a safety office, or a qualification that will expose them to these files, the point of entry is not mastery of the analysis tool: it is the ability to explain, in three sentences and in front of a sceptical crew, what becomes of the data, who can link it to a name, and under what conditions. A programme that no one on the flight deck can explain in those terms produces traces, not safety.

The applicable texts, the operator’s operations manual, its flight data monitoring procedure and the agreement reached with flight crew representatives prevail in all circumstances over this article, whose sole purpose is to shed light on the mechanism.

Questions fréquentes

Is a discrepancy between the trace and the report enough on its own to leave non-punitive processing?

No. The discrepancy is not an established fact, it is an open question: it calls for a context interview, not a change of regime. Only the exit provided for by the written procedure, on the restrictive grounds set out in the texts — an intentional act, a manifest and gross breach of the rules, use of the information for purposes unrelated to safety — takes a case out of routine processing. A discrepancy in duration or order never falls into this on its own.

Should the recording be shown to the crew during the interview?

Yes, but only after the account has been collected, never before. Shown up front, the trace acts as a crutch for memory: the crew fills in the blanks with what the curve suggests, and the independent source is lost. Collected first, the account remains usable, and the later confrontation often brings out a piece of context that neither source carried on its own.

Who should make contact with the crew after a detection?

The designated data holder or an authorised peer analyst, named in the procedure and known to crews. Never the operational chain of command, never a department with a disciplinary purpose, never the person leading an administrative inquiry. The identity of the person making contact is the first signal the crew receives, and it determines the quality of everything that follows.

What happens to the original report once the discrepancy has been explained?

It is kept as it is. It is neither corrected nor replaced: it documents a perception at a given moment, which is precisely the information the recording does not contain. A dated reconciliation note is added alongside it, setting out what the data shows, what the interview brought, and what remains unexplained.

How should an event be handled when no usable data is available?

Exactly like any other, relying solely on the account and on available external sources, such as air traffic control recordings or technical documentation. Varying the standard of proof according to whether a recording exists amounts to instituting two regimes within the same fleet. The coverage gap itself should be logged as a finding of the programme in its own right.

How can you tell whether a discrepancy stems from measurement rather than perception?

By replaying the case through the extraction chain before any contact is made: the nature of the parameter — measured or reconstructed —, the sampling rate, a time reference common to the channels being compared, and the algorithm’s threshold checked against the limit actually applicable to that aircraft in that configuration. This check is short compared with everything it determines, and it closes a significant share of apparent discrepancies.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

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