A crew briefs an RNP approach, compares the displayed position uncertainty to the published accuracy value, and concludes it has a comfortable margin. The reasoning is wrong: the two figures do not answer the same question. The obstacle clearance margin of the final segment was not calculated on the navigation system’s everyday performance, but on a rare event and on the promise that it will be announced in time.
Two statements that look alike but describe different things
A performance-based navigation specification carries two distinct requirements on the flight path, which cockpit usage has merged into one.
The first is an accuracy statement: total system error — the gap between actual position and intended path — stays within the published value for at least 95% of flight time. It describes the usual behaviour of the aircraft-crew pair.
The second is a containment statement, which recent texts prefer to call on-board performance monitoring and alerting: the probability that total error exceeds twice the published value without the crew being warned stays below a very low threshold, on the order of one case per hundred thousand per flight hour. Two elements carry the whole weight: the factor of two, and the no-alert condition.
The two share neither the same mathematical object nor the same unit. The first is a quantile: it locates the body of a distribution. The second bounds its tail and cannot be separated from its detection mechanism. Two systems can share the same 95% quantile with tails several orders of magnitude apart: accuracy therefore does not let you infer containment. Yet that is exactly the inference the cockpit makes when faced with a reassuring uncertainty figure.
Accuracy tells you where the plane usually is, but containment is about the rare cases when it strays far without warning. The system is designed so that if it does stray, the crew will know quickly and take action. Accuracy doesn’t tell you about those rare cases.
On an RNP APCH with an accuracy value of 0.1 NM, the containment statement ensures that the plane won’t exceed 0.2 NM without an alert more than once in 100,000 flight hours. If the plane drifts to 0.25 NM without an alert, that’s a failure of the containment system, not just the accuracy.
What the 95% quantile leaves out
Taking the accuracy statement seriously means accepting that the aircraft spends one twentieth of its time outside the published value. On a final segment that amounts to seconds, and the design has allowed for it: it is not grounds for a go-around.
If the obstacle margin were sized on this statement, the residual risk would be counted in percent per approach. The safety objective adopted for an approach operation sits several orders of magnitude lower — typically on the order of one in ten million per approach, depending on the design framework applied. The gap is not closed by adding margin to the accuracy figure: it is closed by changing the nature of the statement.
Leaving the accuracy value is therefore not dangerous in itself, since the protected area is wider. The dangerous event is leaving it without knowing it: only that is bounded by the containment statement.
Three errors inside one envelope, one automatic watchdog
Total system error breaks down into three terms that share neither the same causes nor the same watchdogs.

- Path definition error: the gap between the path coded in the on-board database and the one the published procedure intended to define. Assumed negligible, this assumption is bought by the data verification chain, not by an on-board sensor.
- Navigation system error: the gap between computed position and actual position. The only term the on-board electronics can bound in real time.
- Flight technical error: the gap between computed position and the displayed path, that is, what the autopilot or the pilot flying lets through. It is monitored via the displayed deviation, hence by the crew.
The on-board monitoring and alerting function therefore bounds only one third of the envelope: it compares an upper bound of position error to an alert limit and ignores how the aircraft is tracking its path. No light comes on because the autopilot is lagging outside in a crosswind.
The design arithmetic then becomes readable. In common frameworks, the integrity monitoring alert limit applies to navigation system error alone and sits on the order of the segment’s accuracy value, whereas the containment envelope applies to total error and equals twice that. The difference is not a comfort margin: it is the budget allocated to flight technical error and path definition. The deviation criterion as usually stated — stay within half the accuracy value, transients tolerated up to the full value coming out of a turn — is the price asked of the crew for the bound to hold true; the binding wording is that of the operator’s documentation.
| What is compared | Accuracy statement | Containment statement |
|---|---|---|
| Question asked | Where does the aircraft usually sit? | How far can it go without anyone knowing? |
| Mathematical object | Quantile of the error distribution | Bound on the tail, coupled to detection |
| Bounded quantity | Total system error | Unannounced total error |
| Reference value | The published accuracy value | Twice that value |
| Probability | One twentieth of time allowed outside the value | On the order of one in a hundred thousand per flight hour |
| Monitored by | No certified real-time function; at best an indicative uncertainty comparison | Integrity monitoring on the navigation side, crew on the flying side |
| Its violation means | A situation anticipated by the design | Loss of the assumption that carries the margin |
| What it sizes | The choice of specification | The area semi-width and the obstacle margin |
Containment is not a distance, it is a bound-alert pair
The word suggests a corridor, and that is the mistake: the aircraft is not confined there by construction. The statement has two inseparable halves, a bound on the probability of excursion and a mechanism that announces the excursion when it occurs. The design assumption is not that the aircraft stays inside, it is that if it leaves, someone knows within seconds and the go-around is flown.
The alert is therefore a load-bearing element of the obstacle margin, just as much as the area width: neutralising it, or delaying the go-around while going to look, removes a piece of the calculation without anything moving on the screens.
This alert is regularly read backwards. It does not say that the aircraft has drifted, but that the system can no longer guarantee its bound: degraded geometry, loss of redundancy, a rejected sensor. Position may be excellent at that very instant. Symmetrically, the absence of an alert does not say that position is good, only that no fault detectable at the required level has been found.
The trap. When manual entry of the accuracy value is permitted, two opposite actions produce the same illusion. A value smaller than the published one narrows no protected area: semi-width and margin were fixed at the design stage. A larger value only pushes the alert threshold further away: the area stays identical, only the sentence that warned you of leaving it disappears. In both cases, the displayed value no longer designates the area actually overflown.
How obstacle design relies on the rare-event statement
The design chain reads in a single direction. The designer selects a specification for the segment, derives from it an area semi-width proportional to the accuracy value — generally twice it — applies an obstacle margin to it, then checks that the residual collision risk stays below the safety objective.
This calculation only holds under three simultaneous assumptions: the total error distribution conforms to the accuracy statement in its body, its tail is bounded by the containment statement, and the alert is followed by a manoeuvre. Remove the third and the safety case collapses, even though the first two remain true. Going around on an integrity alert is already counted in the result that allowed the minima to be published.
Hence a counter-intuitive rule: the difference between the displayed deviation and the area width is not a usable reserve. The result was obtained by integrating over all possible positions; deliberately consuming part of the area does not eat into a margin, it replaces the assumed distribution with another one and invalidates the calculation.
The right reflex. Before reacting to an indication, name the statement it concerns. Growing uncertainty, unfavourable accuracy comparison: everyday statement, dealt with by a flying decision — discontinue, divert, ask for vectors. Integrity alert: rare-event statement, dealt with by the action it was counted for in the design. At the briefing, state two triggers, not one.
Reading cockpit indications: which statement each one answers
Installations display both families of information without labelling them, and some display only one. Comparing an estimated position uncertainty to the required value is an accuracy check, not the monitoring and alerting called for by the specification, which requires an integrity bound paired with a time-to-alert.
| Indication | Statement concerned | What it does not say | Response |
|---|---|---|---|
| Position uncertainty beyond the required value | Accuracy | That the protected area has been breached | Do not commence, or do not continue |
| Integrity alert | Containment, navigation part | That position is bad | The planned action, without visual confirmation |
| Displayed deviation beyond half the accuracy value, no alert | Containment, flying part | That all is well for lack of a light | Correction, then go-around at the briefed criterion |
| Dead-reckoned position after loss of measurements | Neither | That the displayed path remains protected | Exit the specification-based operation |
| Interference notified by notice to airmen | Outside the probabilistic model | That the alert will warn you | Treat the aid as unavailable, independent fallback |
| Entered value different from the published value | Both, misaligned | That the area has changed | Revert to the published value, or discontinue |
The last two rows are the most serious: they remove the validity of both statements without lighting anything up in the cockpit.
What neither statement covers
Both statements are probabilistic and assume random, reasonably independent failures: that is what allows speaking of an hourly rate and then converting it into a per-approach risk. Two families of events escape this model.
The first is systematic data error: a mis-coded waypoint, a wrong cycle, a final segment with a valid seal but false content. It does not show up as noise, it shifts the intended path itself. No position error bound sees it, since the position is correct — it is the reference that is wrong. The only control left on board: cross-check the distance and track of each segment against the chart.
The second is deliberate interference. Jamming removes the signal and gets noticed. Spoofing supplies measurements that are mutually consistent, hence with no detectable residual: integrity monitoring concludes in good faith that there is no fault, and the absence of an alert is the normal behaviour of a system fed consistent lies. The situation falls outside both statements’ assumptions; the remedy lies outside the system — a means of a different nature, consistency with dead reckoning, suspicion of a performance that looks too good.
What to verify in your own documentation
Four points deserve to be found in black and white, since none of them can be read off a chart.
- The nature of the check performed on board for the operation in question: accuracy comparison, an integrity bound paired with a time-to-alert, or both.
- The form of the alert: message, flag, scale change, loss of a mode. Then: can it be inhibited, and what remains displayed afterwards.
- The deviation criterion and its go-around trigger as written in your own documentation, expressed in the displayed scale: a criterion stated in distance does not transpose once the scale becomes angular near the threshold.
- The policy on manually entering the accuracy value: allowed, forbidden, or reserved for named cases — and what must be checked afterwards.
For anyone preparing a qualification, the exercise is not to recite the published value. What is being tested is the ability to say why the area equals twice that value, which term of the envelope nobody monitors on your behalf, and what the alert actually asserts. These three answers form the backbone of the FL330 training programmes devoted to performance-based navigation.
A grey area persists. Converting an hourly rate into a per-approach risk assumes an exposure duration and an independence between approaches that nothing verifies in operation. The budget left for flight technical error is published nowhere: implicit in the factor of two, it says nothing about an admissible deviation compounded with an admissible position error, which the calculation treats as a probability and which the crew encounters all at once. Comparing how each operator translates the alert into action is the subject of the professional discussions.
This article explains mechanisms and sets no operational value. In case of discrepancy, the applicable texts, the charts and notices in force, and the operator’s approved documentation prevail over what is written here.
Questions fréquentes
Why is the protected area twice the published accuracy value?
Because the published value only describes the body of the error distribution, 95% of flight time, and says nothing about its tail. The monitoring and alerting requirement is written on an envelope twice as wide, applying to total system error and paired with a very low probability of unannounced exceedance. It is this rare-event envelope, not the everyday quantile, from which the designer derives the area semi-width and then the obstacle clearance margin.
Does an integrity alert mean the aircraft has left the protected area?
No. It means the system can no longer guarantee that position error stays below its alert limit, which also happens with a perfectly correct position, through simple degradation of satellite geometry or loss of redundancy. The operational consequence is nonetheless the same as if the excursion were real, because the obstacle margin was calculated assuming that this alert would be followed by a manoeuvre. You act on the alert, you do not interpret it.
Does a very small displayed deviation prove that the containment statement holds?
No, because the displayed deviation only informs on one of the envelope’s three terms, flight technical error. It is measured against the computed position, itself tainted by navigation system error, and against a coded path that may differ from the published path. A zero deviation on a mis-coded segment, or with a false position, remains a zero deviation on the screen.
Is the flight management system's accuracy check enough to satisfy the monitoring and alerting requirement?
No, these are two distinct comparisons. Checking an estimated position uncertainty against the required value is an everyday performance test, whereas the monitoring and alerting requirement demands an integrity bound, with a demonstrated confidence level and time-to-alert. An installation may perform the first without the second, and only the approved documentation says which is actually provided for the operation in question.
What are the two statements worth in the case of satellite signal spoofing?
They are worth nothing, because both assume random and reasonably independent failures. Spoofing supplies mutually consistent measurements: integrity monitoring finds no abnormal residual and does not alert, which is its normal behaviour when faced with a false but consistent data set. The only defence lies outside the system, through cross-checking with a means of a different nature and with dead reckoning.
What should be done if the accuracy value displayed on board differs from the one published on the chart?
It must be brought back to the published value before commencing the operation, or the operation must be abandoned. A smaller value does not narrow the protected area, which is fixed at the design stage, and a larger value only pushes the alert threshold further away without widening anything. In both cases the alert stops relating to the area actually overflown, which deprives the obstacle margin of its underlying assumption.

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