Descending towards an intermediate level, the monitoring pilot calls the new clearance, the pilot flying sets the altitude, and the aircraft keeps descending past it. Nothing has failed: the vertical channel is holding a selected path angle, it was never told to level off, and the altitude window was not armed. Mode surprise almost never comes from a system making a mistake; it comes from a gap between what the machine is doing and what the crew believes it is doing. That gap has its own kinematics, its own warning signs, and a time budget set by the aircraft’s energy. It also has a countermeasure, which fits in three questions asked out loud.
The problem isn’t the plane breaking – it’s the pilots and plane not being on the same page. The plane keeps doing what it was told, but the pilots assumed it would do something else. This mismatch grows over time, and the key is catching it early before it becomes dangerous.
The mode annunciator does not say what the aircraft is doing, it says who is flying what
The instruments give a state: a speed, an attitude, a rate. The mode annunciator gives something else, and it is the only piece of information of that kind on board — the machine’sintent . It answers one question per column: who holds thrust, who holds the lateral axis, who holds the vertical axis. It adds a line for what is not yet engaged but will take over next. A pilot who reads a speed knows where they are; a pilot who reads the annunciator knows where they are going.
The distinction sounds academic, yet it explains almost every mode surprise. The crew that gets caught out is usually looking at the right numbers. They see an altitude ticking by, a heading turning, a speed holding, and nothing in those values signals that the vertical channel is not doing what is expected of it. The missing information is not in the parameters: it is in the few words at the top of the screen, which no one has reread since the last action.
A second, more troublesome peculiarity: the annunciator changes on its own. A displayed value stays displayed until you change it; a mode, by contrast, can disappear because an armed condition was met, because a limit was reached, because a tracked signal degraded. It is the only display on the flight deck whose content changes without any hand asking it to, and it is the one the scan visits least often.
On an Airbus A320, an approach mode might disarm automatically if the aircraft drifts more than 1.25 nautical miles from the extended runway centerline. The crew might not notice this change until they see the aircraft continuing to descend without capturing the glideslope, potentially leading to a missed approach situation.
Three families of surprise, three different triggers
Sorting cases into three families avoids the argument over acronyms, which vary from one type to another while the underlying mechanism does not.

First family: the requested mode did not engage. The action was taken, it did not succeed — too brief a press, a condition not met, a mode refused because incompatible with the current state. This case is inherently the mildest, since nothing changes: the aircraft carries on doing what it was doing. It becomes serious when the selection was meant precisely to correct a flight path, because the crew then believes the correction has been made.
Second family: the mode is engaged, the target is not. The word displayed is the one expected, but the value it is obeying is not the one believed: an altitude window left on the previous clearance, a speed computed by the flight management system instead of the selected value, a heading held when the clearance was for a track. The annunciator is innocent; it was only half read, the number next to the word having been overlooked.
Third family: the system changed mode on its own. This is reversion, and it is the only one of the three where the crew did nothing at all. It happens when the condition that made holding possible disappears: an approach signal lost or degraded, leaving the domain where the mode remained valid, a thrust or attitude limit reached, a protection taking over. The system then falls back to a more basic mode — often perfectly logical, and perfectly unexpected.
| Family | Typical trigger | What the annunciator shows | Immediate action |
|---|---|---|---|
| The requested mode did not engage | Press not registered, engagement condition not met, mode refused because incompatible with the current state | The box has not changed, or the expected armed mode does not appear on the bottom row | Redo the selection and check its effect on the parameter before moving on to anything else |
| The mode is engaged, the target is not | Altitude window left on the previous clearance, speed managed by the computer instead of the displayed value, heading held where a track was cleared | The word is correct, the number next to it is not | Reread the target next to the mode, out loud, stating the number |
| The system changed mode on its own | Tracked signal lost or degraded, leaving the holding domain, a limit reached, a protection taking over, an approach mode disarmed | A box changed with no action taken, or an armed mode disappeared without going active | Retake the axis concerned, only then look for why |
The hierarchy of severity is not the one intuition suggests. A selection not taken up is spotted fairly quickly, because an effect was expected and it does not come. A reversion, by contrast, produces stable, coherent behaviour: the aircraft is holding something, just not what was asked of it. No symptom contradicts the mental model as long as the gap stays small, and the gap stays small for exactly as long as it would be easy to fix.
The surprise has kinematics, hence warning signs
Between the moment the gap is born and the moment it costs something, the sequence is always the same. A selection is made, or a reversion occurs. The parameter starts to diverge, but more slowly than the instruments’ reading resolution: for a few seconds, the screen is rigorously consistent with what was expected. Then the gap becomes readable, and it then grows faster and faster, because flight-path errors integrate. Finally comes detection — through a call-out, an alert, or a glance landing in the right place — then recovery.
The practical consequence is unwelcome: the window where correction costs nothing closes before before the parameter betrays anything at all. Basing your vigilance on watching the numbers means structurally accepting that you will arrive in the second half of the timeline. What must be monitored upstream is not the parameter: it is the match between the action, the word displayed and the effect obtained.
The signs that precede the gap are few and recognisable:
- a selection that produced, in the seconds that followed, no change in the targeted parameter;
- an annunciator box that changed although no control was touched;
- an armed mode that disappeared without going active;
- a thrust lever or a trim that moves one way and does not come back;
- two pilots who do not call out the same thing about what should happen next;
- a transition phase — a level change, a reconfiguration, arming an approach, a go-around: the surprise is born there, never in stabilised cruise.
The trap: the silent reversion. Not every mode change comes with an aural alert, and the visual highlighting of a transition lasts a few seconds before fading. If your eyes were elsewhere during that window — on a chart, a frequency, a passenger calling — the information is not delayed, it is lost. The annunciator then settles back into the quiet look of a normal state, and nothing will ever again signal that it is not the one you asked for.
Energy sets the time you have
The clock on a mode surprise is not the mode: it is the energy state. In smooth cruise, far from any limit, a wrong vertical mode buys minutes. In landing configuration, thrust reduced, a few knots above minimum manoeuvring speed, it buys tens of seconds. The same misreading therefore does not carry the same value depending on where in the flight it occurs — and it occurs mainly where time is short, since that is where transitions come thick and fast.
The nastiest configuration is the one where the vertical channel is holding a flight path — a path angle, a plan — while the thrust channel sits in a held-back state where it no longer regulates speed. Nothing has failed, nothing is flashing: the levers stayed where they were left and speed becomes a free variable, a consequence of the imposed flight path. In low-thrust landing configuration, deceleration typically runs at a few knots per handful of seconds; the margin therefore disappears silently, and the first alert arrives only once it has already been used up.
The direction of the error matters too. Climbing, a wrong mode is paid for in speed, meaning quickly. Descending, it is paid for in distance: the problem is slow, almost comfortable, and ends in a high approach caught up in a rushed configuration. Same rule either way: recovery starts with the most constrained axis, and that is almost always speed.
The right reflex: when the flight path is held, call out the speed. The moment a vertical mode is following a plan or a path angle, speed stops being guaranteed by the mode and becomes the parameter to watch. A brief call-out of speed and its trend at each configuration step costs only a few words, and it turns an invisible degradation into shared information.
Detect early: read the words, not the translation
The most common failure mode of monitoring is not the absence of a scan, it is reading by anticipation. The pilot calls out what they expect to see, not what is written; the eye crosses the line and the brain hands back the mental model. It is a sincere, coherent, immediate call-out, and a perfectly wrong one. The other pilot, sharing the same mental model, confirms it without hesitation.
Two habits break the mechanism. The first: state the words in the order they are written, column by column, with the number, never translating them into intent. ‘Path angle, minus one thousand five hundred, altitude armed’ is not the same call-out as ‘we’re descending to the requested level’: the second has verified nothing. The second habit: treat an action as finished only once its effect has been observed, not once the button has been pressed. Between the two lies the engagement condition, which does not belong to the pilot.
That leaves the question of when to reread: constant scanning is unrealistic and no one does it. The useful moments are few — after every action on the guidance panel, at every altitude or configuration change, on arming an approach, and the moment the two call-outs disagree. In other words, during transitions, exactly where workload pushes you not to do it.
Three questions out loud, and one notch less automation
When something does not add up, the temptation is to explain it. You look, you hunt for the parameter that moved, you form a hypothesis — while the aircraft keeps going. The countermeasure fits in three short questions, asked out loud, in this order: which mode is active on the axis concerned, towards which numerical target, and what engages after it.
Their value lies not in the answer but in the time it takes to give it. If either pilot hesitates, or if the two answers differ, the conclusion is already there before any diagnosis: the state of the system is not shared, so it is not under control. A notch of automation is then removed, and the flight path is retaken before hunting for the cause. Diagnosis comes after, with the flight path held, or not at all.
The trap: the half-removed notch. Disconnecting the autopilot while keeping a flight director that still displays the doubtful mode just redoes the mistake by hand, with the same bars and one more pilot in the loop. Either the director’s commands are understood and followed, or they are not and are cleared away to fly on raw parameters, with a called-out target. The in-between state — following bars you cannot explain — is the worst of the three.
What this changes in training and checking
In a session, the examiner does not time the recovery. What they note is the moment of the verbalisation and the consistency of the calls between the two seats. A candidate who brilliantly catches up after a long silence scores worse than one who calls out the doubt almost immediately, removes a notch and stabilises with no particular elegance. The first has shown a skill that does not travel; the second, a method that travels everywhere.
For a pilot arriving on a new type, the trap is vocabulary. Acronyms look alike from one manufacturer to the next without covering the same logic, and the grammar of reversions differs even between two aircraft of the same family. Reading the new annunciator with the old habits produces an understanding that is fluent and wrong — the worst of states, since it triggers no question at all. What genuinely travels is the method: read the boxes, verify by the effect, remove a notch. The rest is learned on the type, and is worked on during preparation for both type rating and selection, where training pathways devote an explicit share to handling automation.
What to go and check in your own documentation
Five points are worth looking up in the flight manual and the operator’s documentation rather than reconstructed from memory. Which reversions does the manufacturer declare, and which come with an aural alert? Under what conditions does an approach mode disarm, and what takes its place? What exactly does the thrust channel do when the vertical channel is holding a flight path, and under what conditions does the speed protection wake it up? How do the protections change according to the level of control-law degradation? Finally, what does the operator’s policy say about the minimum level of automation on approach and about the go-around?
Two grey areas remain, and it is better to know them than to discover them. A single fleet can combine two variants whose reversion logic differs on the point that matters, without the difference being obvious from the flight deck. And training, by its very nature, repeats a finite number of scenarios: crews become excellent at the surprises they know, which says nothing about their behaviour in front of ones they have never seen. It is again the method, not the catalogue, that makes the difference on that day.
This text describes mechanisms, not procedures: the applicable regulation, the manufacturer’s flight manual and the operator’s approved documentation take precedence in all circumstances over what is written here, and it is that documentation that governs before the authority as before the crew.
Questions fréquentes
What is the difference between an armed mode and an active mode?
An active mode is actually holding the axis concerned right now; an armed mode is waiting for an engagement condition before taking its place. The switch from one to the other happens with no crew action at all, the moment the condition is met, and the mode it replaces then disappears from the annunciator. That is why the armed-mode row must be read on the same footing as the active-mode row: it announces the behaviour of the minutes to come.
Why is a reversion more dangerous than a failed selection?
Because it contradicts nothing the crew observes. A failed selection gives itself away quickly, since an effect was expected and it never comes. A reversion, by contrast, leaves the aircraft holding something clean and steady, so no symptom clashes with the mental model as long as the flight-path deviation stays below the instruments’ reading resolution. On top of that, it happens precisely when an outside condition has degraded, that is, exactly when workload is already high.
How do you really check that a mode has taken, rather than just reading it?
By treating an action as finished only once its effect has been observed, not once the button has been pressed. After every selection, check in the following seconds that the targeted parameter is moving in the right direction at a plausible rate. Reading the displayed word confirms the system’s intent, the parameter’s movement confirms that it is executing it: both are needed.
What does removing a notch of automation actually mean?
It means stepping down to a level where you know exactly what is commanding the flight path, and doing so wholly. Disconnecting the autopilot while still following a flight director that displays the doubtful mode removes nothing: the error is simply reproduced by hand. Either the director’s commands are understood and followed, or they are cleared away and the aircraft is flown on parameters with a target called out loud.
Why do mode surprises mainly occur during transitions?
Because a transition combines all three factors: many selections in a short time, workload that pulls the eyes away from the annunciator, and engagement conditions changing from one moment to the next. In stabilised cruise none of these factors is present and the annunciator stays unchanged for hours. The moments worth rereading are therefore level changes, reconfigurations, arming an approach and the go-around.
Aren't envelope protections enough to catch a mode surprise?
They are a last net, not a form of monitoring. A protection steps in once a parameter hits a limit, that is, after the margin has already been used up, and it often makes recovery costlier in altitude than if it had been triggered a few seconds earlier. Moreover, their availability depends on the level of control-law degradation, information found in the flight manual that must be known before relying on it.

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