The Tablet Cannot Tell You Its Data Is Wrong

A pilot in a cockpit at dusk, examining a tablet displaying flight data with no visible distinction between correct and incorrect information.

A navigation database and a takeoff performance figure land on the same screen, look equally authoritative, and come from entirely different places. Between the State that published the raw information and the tablet clipped beside the window there are several organisations, half a dozen file formats and at least one network. Every handover is a point where a value can change, and almost none of them announce that anything has.

An electronic flight bag is a display device with a processor attached. Whether it is a portable unit in a mount or an installed one wired into the aircraft, its defining property is that it holds no independent knowledge of the aeroplane it is sitting in. It does not measure mass, it does not sense the runway, it does not know which engine standard is fitted unless somebody configured it to know. It renders what it was handed, with the same crisp typography for a correct figure and a wrong one.

Regulators split the software on it by consequence rather than by function. The lightest category covers applications whose failure or misuse has no operational effect worth managing; the next covers those that do, and that therefore need an assessment, a configuration baseline and a place in the operator’s procedures. Performance and mass-and-balance applications sit firmly in the second group, and so do the chart and airport moving map applications that crews now use in place of paper. The classification matters because it decides how much process stands behind the file, not how convincing it looks on screen.

Two very different kinds of payload arrive by that route. One is static data prepared elsewhere and shipped: navigation databases, terrain and obstacle sets, charts, airport data. The other is a computation performed on the spot from certified tables embedded in the application, with inputs typed or fetched at the moment of use. They fail in different ways, they are protected by different mechanisms, and confusing them is the fastest way to trust the wrong safeguard.

The tablet shows data, but it doesn’t know if that data is right or wrong. It just displays what it’s given, so pilots must check carefully.

If a tablet shows a runway length of 2,500 meters when it’s actually 2,800 meters, the pilots won’t notice unless they cross-check with another source. The tablet won’t warn them.

What stands behind a navigation database

Aeronautical data starts at a State authority, which publishes it in its aeronautical information publication under the internationally fixed regulation and control cycle — the twenty-eight day rhythm that puts amendments into force simultaneously worldwide, at a published effective time. That synchronisation is the reason a database has a cycle number at all, and the reason an out-of-cycle correction is an event rather than a routine.

Three people in an office review a flight map on a screen while one points to a paper chart.
Cross-checking digital and paper data ensures accuracy in critical systems.

From there the information passes to a data house, which is where the most under-appreciated work happens. Publications from a large number of States, in different formats and different levels of precision, are aggregated and coded into a machine-readable structure. Coding a procedure is not transcription: it is interpretation. A published departure has to be expressed as a sequence of legs and terminators that an avionics system will fly, and the choices made there decide what the aircraft actually does. Two suppliers reading the same national publication can produce two defensible codings that behave differently in the air.

The package is then tailored for the customer — company routes, operator-specific waypoints, restricted content removed or added — and built into deliverables. This is where a single cycle stops being a single thing. The load that goes into the flight management system, the chart set that goes onto the tablet, the terrain database and the airport map are separate builds from a common source, produced on different schedules and installed by different means. They can and do diverge, and nothing on the aeroplane compares them.

Distribution is the last leg: a supplier portal, an operator’s device management system, an aircraft network, a hotel wireless connection. Then installation on a device that may have been offline for a fortnight.

The chain behind a database, and what each link can change
Link Who owns it What can go wrong there What would reveal it
Source publication State aeronautical information service An error in the published data itself; a late amendment Nothing downstream — the error is authoritative
Coding and aggregation Data supplier Procedure coded in a way that flies differently from the published intent Comparison of the flown path with the chart, in the aircraft
Operator tailoring Operator plus supplier Stale company routes, obsolete tailored waypoints Periodic review; crew reports
Packaging per target Data supplier Chart set and avionics load out of step for the same cycle Deliberate comparison of cycle identifiers, device by device
Distribution Operator IT and device management Partial download, failed install, device offline past the effective date The device’s own currency indication, if anyone reads it
Rendering Application on the device Correct data displayed at the wrong scale, zoom or filter setting Cross-reading between crew members

A performance figure has no chain — it has a configuration

A takeoff or landing performance result is not shipped. It is produced when the button is pressed, from three ingredients: a certified performance core derived from the aircraft flight manual and embedded in the application, an aircraft configuration file that describes this specific tail number, and inputs supplied at the moment of use.

The certified core is usually the most solid part of the whole arrangement. It carries a version, it was validated against the flight manual, and it changes rarely. The configuration file is a different matter: it encodes engine standard, thrust rating, aerodynamic modifications, installed options and the effect of any deferred defect that alters performance. It is maintained by the operator, not the manufacturer, and it drifts as the fleet drifts — a tail number that received a modification, an engine change, a winglet retrofit. A configuration file that lags behind the aeroplane produces a result that is internally consistent, plausible and wrong.

The inputs are the third ingredient, and they arrive by increasingly automated paths — a loadsheet by datalink, a runway condition report, a meteorological observation, a runway selection made from the airport database. Automation removes transcription errors and introduces a subtler one: an input that is wrong at the source now propagates without a human ever looking at it.

Performance numbers are calculated on the spot, but if the inputs or settings are wrong, the result can be dangerously off without any warning.

If a Boeing 737’s configuration file doesn’t account for new winglets, the takeoff performance calculation might underestimate the required runway length, leading to a dangerous situation.

What a checksum proves, and what it cannot

Aeronautical data is governed by a quality framework that describes each item by accuracy, resolution, integrity, traceability and timeliness, and classifies it by the consequence of corruption — routine, essential, critical. The critical class carries a target for undetected corruption several orders of magnitude more demanding than the routine one, which is why the more consequential classes travel with cyclic redundancy protection and why the processing organisations that handle them work to a recognised data processing standard and hold a formal acceptance from an authority. In the United States that acceptance takes the form of a letter of acceptance issued to the data supplier; in Europe, data service providers hold a certification of their own, granted under a dedicated set of rules.

All of that protects the journey. None of it protects the meaning. A cyclic redundancy check confirms that the bits that arrived are the bits that left. A digital signature confirms that the file came from the organisation that signed it and has not been altered since. Neither says the content was correct when it was signed, nor that this file is the right one for this aircraft, this operator or this cycle. An error introduced upstream is carried faithfully, sealed and verified, all the way to the screen.

Protection mechanisms and their blind spots
Mechanism Catches Does not catch
Cyclic redundancy check on the data set Corruption in storage or transfer An error present before the check was computed
Digital signature on the package Substitution, tampering after issue, unknown origin A genuine package that is the wrong build or the wrong cycle
Supplier acceptance or certification Absence of a documented process at the data house A defensible but incorrect coding decision
Currency indication on the device An expired cycle, when the crew reads it A current cycle that was built from bad source data
Application input validation Values outside plausible ranges, missing fields A plausible wrong number, which is the dangerous kind

When two crew members stop being two sources

The historical defence against a bad number was independence: two people, two calculations, two documents, one comparison. Every element of that has quietly eroded. Two tablets running the same application, loaded from the same distribution, using the same aircraft configuration file, fed the same datalink loadsheet, are not two sources. They are one source with two screens. When the crew then reads the inputs across to each other rather than entering them separately, the comparison confirms transcription and nothing else.

The consequences are documented. The most instructive case remains the 2009 departure from Melbourne of a long-range four-engined airliner, investigated by the Australian authority, in which a takeoff weight roughly one hundred tonnes below the actual mass was used. Everything downstream of that figure was consistent with it: the thrust setting, the speeds, the cross-check. The aeroplane struck its tail and passed beyond the end of the runway before flying. The error was not detected because there was nothing in the loop capable of disagreeing with it.

Restoring independence does not require going back to paper. It requires at least one check whose input path is genuinely different: a gross error check against a memorised or tabulated expectation, a comparison with a figure computed by the flight management system from its own database, entry of the critical inputs separately by each crew member rather than by dictation, or a sanity relation the crew knows by heart between mass, flap setting and speed.

The reflex worth having. Before accepting a performance result, ask what would have had to be different for the answer to come out wrong and still look like this. If the honest answer is a single input, a single file or a single distribution, the cross-check performed was not a cross-check.

The device, its custody and its connections

A portable electronic flight bag spends most of its life away from the aircraft. It is carried through hotels, connected to networks nobody controls, updated by an operating system vendor on a schedule set in another industry, and in some operations owned by the crew member rather than the company. That is a custody problem before it is a security problem.

The practical failure modes are mundane. An operating system update changes a permission and an application stops reaching its data. An application auto-updates outside the operator’s approved version list. A device management enrolment expires and the device silently stops receiving pushed content. A tablet used only occasionally arrives at the aircraft with a cycle that ended a week ago. A shared credential means nobody can say which device received which build. None of these require an adversary; all of them produce a device that is out of configuration while looking normal.

The deliberate cases are rarer but structurally similar, and they are the reason information security has moved from the technical annex into the management system. European rules now oblige aviation organisations to manage information security risk with the same formality as safety risk, which forces a question that used to fall between departments: who could alter this data set, and would anyone notice?

The trap. A data set that is identical across the whole fleet is not evidence that it is correct. It is evidence that one distribution worked. Uniformity removes the very disagreement that used to expose a bad build, which is why a corrupted upstream package is the failure mode with the widest reach and the quietest signature.

The organisation standing behind the screen

Every operator running an electronic flight bag has to name someone responsible for it. That role — administrator, by whatever local title — owns the configuration baseline: which hardware, which operating system versions, which applications at which release, which data sets, and how each of them is verified as current. It also owns the part that gets designed last and needed first, which is withdrawal. Pushing a corrected build is straightforward. Ensuring that two hundred devices have actually taken it before their next departure, and knowing which ones have not, is an operational problem with a deadline.

The safety management obligations sit on top of that. A change of version, a change of supplier, a move from paper to screen or from one device family to another is a change to be assessed before it is made, with the hazards identified and the mitigations decided in advance. In practice the assessment tends to concentrate on the transition and to say little about steady-state data integrity, which is where the residual risk actually lives.

What remains in the crew’s hands

Very little of this chain is visible from the flight deck, but the visible part is not nothing. The cycle identifier and effective date can be read rather than assumed, on each device and for each data set rather than once for the tablet as a whole. The aircraft configuration selected in a performance application can be checked against the tail number actually being flown, including engine standard and any performance-affecting deferred item. A result can be weighed against an expectation before it is used. And where a genuinely independent figure exists, it can be compared instead of admired.

The last one matters beyond the flight concerned. An anomaly noticed and reported after an uneventful sector is the only mechanism by which an upstream error becomes visible at all, because the same package is sitting on every other tablet in the fleet, behaving exactly as consistently. Crews comparing what they see across operators and types is one of the few places these divergences surface early, and it is a recurring thread on forums.fl330.com.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

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