Safety recommendation: its lifecycle, the authority’s response, and the meaning of the final classification

A senior investigator reviews a final safety report in a dimly lit office, with a timeline diagram of the recommendation lifecycle projected on the wall.

A final report is issued, containing several safety recommendations. A few years later, the tracking table shows all of them closed, with the response deemed satisfactory, and the crew flying the affected type has noticed no changes in their documentation or the standard of their aircraft. There is no procedural anomaly here: the classification is based on the recipient’s response, never on the state of the risk. Everything else follows from this distinction.

A recommendation does not obligate anything, except to respond

The starting point is legal and still surprises experienced practitioners: a safety recommendation creates no obligation to act. It comes from a safety investigation authority, i.e., an organization whose mission is prevention, not punishment, and it is addressed to an entity with the power to change something. What it creates is an obligation to respond: to acknowledge receipt, then to state what has been done, what is planned, and by what deadline, or why nothing will be done.

This asymmetry is intentional. An investigation authority that could impose a design modification would become a second regulator, and the independence of the investigation—the kind that allows testimonies to be collected without fear of disciplinary action—would not survive. The price of this independence is that the recommendation has only the force of its demonstration. It only becomes binding if another actor adopts it: the regulator through an airworthiness directive or rule amendment, the operator through an approved modification of its documentation.

Life cycle of a safety recommendationTimeline with five successive markers: occurrence, investigation, final report, addressee’s response, closure. A dashed branch leaves the investigation and goes straight to the response, representing a recommendation notified before the report is published. A bracket below the timeline covers the period from the final report to closure, during which the investigation authority actively follows the file.immediate recommendationoccurrenceinvestigationrapport finalresponseclassementdeadline set by the regulationactive follow-up of the filetemps
Follow-up does not begin with the publication of the report. A recommendation may be notified during the investigation, and its response already closed by the time the report appears. Closure closes the bracket; it does not close the subject.

Une recommandation de sécurité n’oblige pas à agir, mais à répondre. L’autorité d’enquête vérifie seulement si le destinataire a expliqué ce qu’il compte faire, pas si l’action réduit vraiment le risque. C’est comme un conseil médical : le médecin peut recommander un traitement, mais c’est au patient de décider s’il le suit.

Après un incident sur un Airbus A320 en 2020, l’autorité d’enquête a recommandé à Airbus de modifier la formation des pilotes sur un système de freinage. Airbus a répondu qu’il mettrait à jour son manuel de formation dans les 6 mois. Le dossier a été classé “réponse adéquate”, même si certains pilotes n’ont vu la modification que 2 ans plus tard, car le suivi administratif ne vérifie pas l’application réelle.

The timeline matters as much as the content. A recommendation does not wait for the final report to be published: as soon as the investigation identifies a hazard requiring immediate action, it is notified separately. Monitoring then begins very early, sometimes just a few weeks after the event, and the recipient’s response may already be classified by the time the report is published. Reading a final report in the belief of discovering new recommendations often leads to rediscovering an action already settled elsewhere.

The recipient is chosen for their ability to act, not for their responsibility

There is nothing preventing a recommendation from being addressed to an operator, but it is rare and almost always a sign of an upstream failure. An event that could only occur with one operator falls under the oversight and safety management system of that operator; it does not call for a public recommendation. When the investigation authority issues one, it is because they have identified a generic mechanism: an ambiguity present in an entire manufacturer’s documentation, a system behavior not covered by a type’s training program, a gap in a reference. The recipient is then the one who can address the entire exposed population: the certification agency, the design approval holder, a state authority, a training organization.

Second rule of drafting, often misunderstood: a recommendation must neither name a culprit nor presume responsibility. This is not a diplomatic precaution, it is a survival condition for the system: the day a recommendation is used as evidence, witnesses stop talking. In practice, this results in formulations focused on the failed barrier, not on the person’s action—and this explains why some recommendations seem lukewarm upon reading, even though they target a perfectly identified hard point.

The response: what the recipient must produce

The recipient has a short deadline—on the order of a quarter in international and European frameworks—to make their position known. Three outcomes are acceptable, one is not. They can describe an action already underway, with its nature and scope. They can announce a future action, and that is where everything is at stake: an intention without a deadline or verifiable milestone is not an action. They can also refuse, provided they justify it—a well-argued refusal, supported by data, is a full and complete response, often more useful than a superficial agreement. What is not acceptable is silence.

Acceptable outcomes of an addressee’s responseDecision sequence. The notified recommendation raises a question: is action possible? If so, the addressee announces a dated, verifiable action; if not, it gives a reasoned refusal. Both branches lead to assessment and then closure. A dashed loop returns to the recommendation: new information may reopen the file.recommendationnotifiedactionpossible ?ouinonaction, datedand verifiablerefusal, reasonedand arguedassessmentthen classificationnew information: reopening
Three outcomes are acceptable; silence is not one of them. A reasoned refusal is a full and complete response, often more useful than an agreement without a milestone. The reopening loop is only triggered if someone brings new information.

Quand une autorité d’enquête envoie une recommandation, le destinataire doit répondre rapidement. Il peut soit dire qu’il va agir (avec une date précise), soit refuser (en expliquant pourquoi). Ce qui compte, c’est que la réponse soit claire, pas forcément que l’action soit parfaite. Si le destinataire ne répond pas, c’est un problème.

Après un incident sur un Boeing 737 MAX, l’autorité d’enquête a recommandé à Boeing de modifier la formation des pilotes sur le système MCAS. Boeing a répondu qu’il allait publier une mise à jour dans les 3 mois, mais sans donner de date précise. L’autorité a classé la réponse comme “partiellement adéquate” et a demandé plus de détails, montrant qu’une réponse vague n’est pas acceptable.

A solid response is recognized by three markers. First, an explicit scope: which aircraft, which organizations, which operations. Second, a dated and externally verifiable milestone—publication of a bulletin, an amendment, an information note—because a milestone internal to the recipient cannot be verified. Finally, a way to verify the effect, and this is almost always missing, even in responses classified as adequate: the system knows what it has decided, but it is much less able to verify that the decision produced the expected result.

What the investigation authority examines therefore is not goodwill but coverage. An action that only addresses the exact configuration encountered during the event leaves the generic mechanism that the recommendation targeted intact; this is the most frequent reason for partial classification. An action deferred to a study without a dated decision point has the same effect: the file remains open, or it closes with a note of disagreement.

The right reflex. Never read a recommendation without reading the response it received and the reason for its classification. The recommendation text states what an investigator deemed necessary; the response states what the system agreed to do, and it is this second piece of information that determines what will eventually reach your aircraft.

Classification crosses two axes, and the eye only reads one

The vocabulary varies from one authority to another, and that is the first source of confusion. The structure, however, does not vary: all classifications cross two independent pieces of information. On one side, the judgment on the response—adequate, partially adequate, inadequate, or assumed disagreement. On the other, the administrative status of the file—open, therefore still monitored and relatable, or closed, therefore out of the loop.

Two men in an office review documents and a computer screen with a list of closed safety recommendations.
Closed safety recommendations require thorough review and classification.
Closure matrix: assessment of the response and status of the fileTwo-way matrix. Rows: the assessment of the response — adequate, partial, inadequate. Columns: the status of the file — open or closed. The shaded closed-files column has three cells, including one that combines an inadequate response with closure on recorded disagreement.assessment of the responsestatus of the fileouvertclosadequatepartielleinadequateimplementation in progressresponse acceptedfurther information requestedclosed with a gapfollow-up or reminderdisagreement recorded
Closure combines two independent pieces of information, and the eye retains only one. Three of the six cells read “closed”, including the one where the described risk received no response judged sufficient. The vocabulary varies between authorities; the structure does not.

Quand un dossier de recommandation est classé, deux choses sont vérifiées : 1) Si la réponse était bonne (complète, partielle, ou nulle), 2) Si le dossier est encore ouvert ou fermé. Beaucoup regardent seulement si le dossier est fermé, mais c’est la combinaison des deux qui compte vraiment.

Après un incident sur un Embraer E-Jet, l’autorité a recommandé une modification de formation. L’opérateur a répondu qu’il allait mettre à jour ses manuels, mais sans date précise. Le dossier a été classé “partiellement adéquate, dossier ouvert”, montrant que la réponse n’était pas complète mais que l’autorité continuait à suivre le dossier.

The box that traps everyone associates an insufficiently judged response with a closed file. There is nothing absurd about it: when the recipient has stated their reasons, the investigation authority has maintained their position, and no new elements are expected, prolonging the exchange adds nothing. The file closes on an acknowledged disagreement, publicly visible, and the described risk remains exactly where it was. Two other reasons for closure have no more to do with the treatment of the danger: the recommendation replaced by a more precise one, and the recommendation becoming irrelevant because the affected fleet has been withdrawn or the organization dissolved.

What a closed file really says, depending on the reason for closure
Reason for closure What is established Typical residual risk What to check in your organisation
Adequate response, action implemented The addressee has acted and the investigation authority has judged it sufficient Decreasing, as the fleet is brought into compliance The scope: type design, serial numbers and standards covered
Adequate response, action announced A dated commitment has been accepted Unchanged until the milestone is passed The announced date, and what has been published since
Partial response accepted Part of the mechanism is addressed The rest of the mechanism, which the closure does not describe What the action does not cover, to be addressed through procedures
Disagreement recorded The positions are set out and the exchange is exhausted The described risk, in full An internal assessment, as for a hazard identified in your own organisation
Recommendation superseded A more precise wording has taken over Carried over to the successor file The reference of the superseding recommendation
No longer applicable The targeted population no longer exists at the addressee Nil for them, unchanged for you if you still operate the type Whether the equipment concerned is actually in your fleet

A closed file is not sealed, however. A new element—a similar event, an operational data point, a subsequent investigation that intersects with the same mechanism—can lead to reopening the monitoring. In practice, the investigation authority more often issues a new recommendation that replaces the previous one. Seeing the same recommendation return a few years later in a similar formulation is in itself a first-order piece of information: it indicates that a mechanism is resisting treatment.

The trap. A tracking table filtered only by the « status » column gives a false and reassuring image. Three closed recommendations may cover an action actually deployed, a final disagreement, and a fleet withdrawn from service by another operator—three situations of which only one concerns you, and it is not necessarily the one you think. The useful filter is on the reason for closure.

Why closure precedes the effect by several years

Even in the favorable case—adequate response, real action, closed file—the risk does not disappear at the time of classification. It then decreases, slowly, at the pace of the technical chain that carries the action to the aircraft in service. The classification marks the end of administrative monitoring; the penetration of the measure into the fleet is a distinct phenomenon, with its own time constant.

End of administrative follow-up and decline of the real riskTwo curves on the same time axis. The dashed line represents the attention paid to the file: it stays level then drops sharply at closure. The solid line represents the fleet’s actual exposure: it declines slowly and continues well beyond closure. The area under this line after closure represents the residual risk.risk levelfile follow-upfleet exposureresidual riskclassementtemps
Closure drops follow-up to zero; exposure declines over years. The area under the solid line after closure is borne entirely by operators: nobody chases any more, and the measure has not yet reached all the aircraft.

Quand un dossier de recommandation est classé, cela ne veut pas dire que le risque a disparu. Le suivi administratif s’arrête, mais il faut des années pour que la mesure atteigne tous les avions. C’est comme un médicament : le médecin peut dire que le traitement est terminé, mais il faut du temps pour que le corps soit complètement guéri.

Après un incident sur un ATR 72, l’autorité a recommandé une modification de formation. L’opérateur a répondu qu’il allait mettre à jour ses manuels, et le dossier a été classé “réponse adéquate” 6 mois plus tard. Pourtant, il a fallu 2 ans pour que toutes les formations soient mises à jour, et certains pilotes ont continué à utiliser l’ancienne procédure pendant cette période.

Two paths go through the technical world, and their orders of magnitude are not comparable. The first goes through design: finding, study, modification, service bulletin, then airworthiness directive if the authority deems the condition unsafe, with its compliance deadline—and the effect only touches the serial numbers and standards concerned. The second goes through the certification reference: a regulatory task is opened, the specification is amended, and the amended text applies to new type definitions. The existing fleet, certified on a previous basis, is not affected. This is the least understood mechanism in the entire monitoring process: a recommendation can be classified as satisfactory because a specification has been tightened, without a single aircraft in service being modified.

Orders of magnitude of timescales, from file to fleetFour horizontal bars on an axis graduated in weeks, months and years. The addressee’s response is counted in weeks, assessment and closure in months. The airworthiness directive and its compliance period, then the amendment of a certification specification, are counted in years. A dashed line marks the moment the file is closed, located after the first two bars only.closureresponseclassificationinstructionspecificationsemainesmoisyears
The first two bars occur before closure, the next two after. The file closes at the pace of a written exchange; design and the certification basis move at the pace of an industrial cycle. The scale is indicative; the difference in order of magnitude is not.

The third path remains, the shortest and the only one the operator fully controls: procedures. An amendment to the operations manual, an evolution of briefings, a point added to the training program produces an effect in weeks where design takes years. It is also the path that does not appear in any recommendation tracking table, because it does not fall under any official recipient—and because no one outside will come to verify that it has been taken.

Recommendation, finding, directive: three objects that are confused

The confusion of vocabulary has immediate operational consequences. A safety manager who treats a recommendation as an audit finding mobilizes resources on an object that does not oblige them; the one who treats a finding as a recommendation misses a deadline and exposes themselves to an escalation on their approval. The three instruments are distinguished on five points, and the most discriminating is not the binding force but what closes the file.

Three instruments that are often confused, and what separates them
Criterion Safety recommendation Oversight finding Airworthiness directive
Émetteur Safety investigation authority Authority overseeing the organisation Authority responsible for the type design
What it requires A response, within a set deadline Correction, within a set deadline, or escalation Compliance, as a condition of continuing airworthiness
Destinataire Whoever can act on the whole exposed population The audited organisation, and it alone Every holder of an aircraft within the applicability
What closes the file A response judged acceptable A corrective action accepted then verified Actual, traceable compliance
Effect on your fleet Indirect, delayed, sometimes nil Direct on your organisation, not on the type Direct and dated, on the aircraft concerned

The recommendation nevertheless retains a value of anticipation that the other two do not have. It appears before the directive, sometimes years before, and it describes a mechanism rather than imposing an action. An operator who reads the recommendations addressed to the type certificate holder of their fleet knows what could become mandatory, and can build their operational defense while the file is still open—instead of suffering it with an imposed compliance deadline.

What to verify at home once the file is closed elsewhere

Three checks transform passive reading into useful work. The first concerns applicability: does the recommendation affect your type definition, your serial numbers, your avionics standard, your operating domain? A corrective action that only targets certain standards leaves the rest of the fleet at the initial risk level, and this information is not in the classification but in the service bulletin. The second concerns local translation: if the announced action is an evolution of documentation, it must have arrived in your manual, your minimum equipment list, or your training program—and if it is not there, no one will tell you, since the file is closed. The third concerns recommendations closed in disagreement: these are the only ones for which we know, in black and white, that the described risk has not been addressed. They deserve an internal evaluation, exactly like a hazard identified by your own system.

A gray area remains, and it is better to name it than to discover it. A recommendation addressed to another state’s authority may be classified as satisfactory without any effect crossing the border, even though the same type flies with you. A recommendation classified solely on the opening of a regulatory task may remain without concrete translation for an entire amendment cycle. In both cases, the tracking table is accurate and the safety gap remains: this is precisely where an operator’s safety management system distinguishes itself from a compliance binder.

For those preparing a type rating, these files are also a profitable read: they designate, better than any manual, the known hard points of the machine and the scenarios that examiners favor. Exchanges between operators on the FL330 forum regularly show that the same recommendation has been translated in three different ways depending on the airlines, for identical fleets.

The applicable texts, the decisions of the competent authority, and the approved documentation of your operator take precedence over this article, which describes a general mechanism and does not replace a safety analysis or expert advice.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

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