Reading a final investigation report: the reading order that avoids premature conclusions

Rapport technique ouvert sur une table en acier, réglette et loupe posées à côté sous lumière froide

The final report is published on a Tuesday. By noon, the preparation room has decided: “they should never have continued the descent.” No one went further than the summary and the list of recommendations—the only two parts of the document that do not establish anything by themselves. A report follows a prescribed order of writing; it is not read in that order, and the discrepancy determines what is taken from it.

An investigation report is not a verdict, and its structure reflects that.

The safety investigation has a single objective: to understand in order to prevent. It does not assign fault or responsibility and runs parallel to judicial or administrative procedures, which seek to identify a responsible party. Both investigations focus on the same flight, often rely on the same evidence, and may produce irreconcilable texts without either being wrong. Reading a final report as a verdict is applying a different framework to the document.

The standardized structure encodes this intent, which is why it is useful. Basic information gathers what is established, dated, and sourced—flight progression, personnel, aircraft, weather, recordings, wreckage, tests, and especially information on organizations and management. The analysis presents the investigator’s reasoning, including hypotheses and uncertainties. Conclusions separate findings from causes and contributing factors. Recommendations are addressed to named third parties. Annexes contain the raw material.

The gradient is clear: from the first part to the last, one moves from established facts to interpretations, then to demands. Starting with the summary means taking the interpretation before the facts; these are no longer read as facts but as illustrations of an already established conclusion, making disagreement impossible to express due to the lack of understanding of the reasoning’s foundation.

A safety report is not about blame—it’s about learning. The structure is built to show facts first, then analysis, then suggestions. If you jump to the summary, you’re guessing the ending before seeing the movie.

Order of the report chapters and recommended reading orderTwo columns of six headings linked by crossing dashed lines. On the left, the page order: synopsis, factual information, analysis, conclusions, recommendations, appendices. On the right, the recommended order: factual information, appendices, analysis, conclusions, recommendations, and the synopsis last.document orderreading ordersynopsisfactual informationanalysisconclusionsrecommendationsappendicesfactual informationappendicesanalysisconclusionsrecommendationssynopsis
The synopsis is read last, as a check, never as an introduction. The two chapters that move furthest are those that ordinary reading treats the wrong way round: the synopsis, which moves down, and the appendices, which move up.

Knowing what document you have in front of you.

Several documents circulate under the name of “report,” without having the same status. The preliminary report, published shortly after the event, presents facts without conclusions: it describes what was observed, not what explains it. Interim information may follow if the investigation is lengthy. A safety recommendation can be issued at any time, as soon as the authority deems that action cannot wait: it addresses a specific point, not the entire investigation.

After the 2018 Lion Air 610 crash, a preliminary report noted the MCAS activation but did not yet establish its cause. The final report, published later, identified the design flaw and led to Boeing’s 737 MAX grounding.

The successive documents of an investigation and what each containsA horizontal time arrow carries four milestones: the occurrence, the preliminary report containing only facts, the draft report submitted to the parties whose comments are appended, then the final report, the only document to carry conclusions and recommendations. A vertical arrow from the middle of the timeline recalls that an urgent recommendation may be issued at any time.urgent recommendation possiblepreliminary reportfaits seulsrapport finalconclusions and recommendationsoccurrencedraft to the partiescomments appendedtemps
Only one of these documents concludes; the others merely establish. Quoting a preliminary report as if it identified a cause, or an urgent recommendation as if it summed up the investigation, gives the status of a conclusion to a text that has none.

Before publication, the draft final report is submitted for comments to the concerned states and parties involved in the investigation. Received comments are either incorporated or appended to the document. This chapter is the most instructive and the least read: when a manufacturer or authority disputes a formulation, it identifies the exact point where the establishment of facts remains debated. This provides a map of the fragile areas, drawn by those who had the most interest in knowing them.

The reading order that protects against hindsight.

The reader of a report has, from the cover page, the only piece of information the crew never had: the outcome. Each decision of the flight is therefore judged in the light of a result that could not have been known at the time. This is not a lack of attention but a mechanical effect of position, and it is not corrected by goodwill: it is corrected by the order in which the chapters are opened.

Information available to the report reader and to the crew as the flight unfoldsOn a horizontal axis representing the course of the flight, a high solid straight line represents the reader’s information, at its maximum from the first page since the reader knows the outcome. A dashed stepped curve represents the crew’s, which starts very low and rises only in steps. The vertical gap between the two is hindsight bias.information disponiblecourse of the flightwhat the reader knowswhat the crew knewhindsight gapa moment in the flight
From the title page, the reader holds the one piece of information the crew never had: the outcome. It is this gap, not a difference in competence, that makes the in-flight decisions look so obvious from a distance.

The sequence that follows takes more time than ordinary reading and yields a completely different result.

  1. The flight progression alone, stopping before the paragraph where the outcome appears, with the timeline reconstructed on a sheet.
  2. Information on the aircraft, crew, operator, and organizations: the most tedious part, and the one that provides the context without which no decision is understandable.
  3. The annexes, with plots and transcriptions first, before any synthesis.
  4. The analysis, identifying at each step the evidence it relies on.
  5. The findings, followed only then by the causes and contributing factors.
  6. The recommendations, with their addressee.
  7. The summary last, as a check: does it say what the body of the document establishes?

Good reflex: read the flight progression while marking each decision point and noting what the crew had at that moment—what they saw, what they received, what their documentation prescribed. Where you cannot write anything, the report does not establish it either: it is a question to ask, not a blank to fill with assumptions.

Read the annexes before the synthesis.

The annexes contain what the body of the report summarizes, and a summary erases uncertainties. A phonetic transcription opens with its conventions: inaudibility, uncertain attributions, precision of timestamps, and the scope of the sequences retained—all of which change the meaning of everything that follows. A plot is not a continuous observation: each parameter has its sampling frequency, resolution, and sometimes a calculated rather than measured origin.

A peak shorter than the sampling interval does not appear in the traceA solid curve shows the real value of a parameter: low, it rises sharply to a narrow peak then falls again. Dashed vertical lines mark the sampling instants; the recorded points, joined by a dashed line, bracket the peak without reaching it, so the recorded maximum remains well below the real maximum.parameter valuetime, one point per samplereal valuerecorded maximumgap
A trace does not show what happened; it shows what was sampled. Sampling rate, resolution and computation method are given in the appendix; a maximum read from a summary plot is a lower bound, never a measurement.

The same caution applies to reconstructions. An engineering simulation does not compare a model to reality but to another model, the one considered representative; its value depends entirely on the input hypotheses written in the annex, not in the analysis. A ground test reproduces a condition, never all the conditions of the day. The investigator knows this and writes it; a synthesis sentence cannot carry all these reservations.

The pitfall: treating a value read on a synthesis curve as a measurement. A peak shorter than the sampling interval does not appear, and a reconstructed quantity inherits the uncertainty of everything used to calculate it. Presenting an argument from a report to a safety commission based on such a maximum is asserting more than the report asserts.

Finding, contributing factor, cause: three statuses often confused.

The conclusions section is not a list of charges ranked by severity. It separates statements that do not have the same scope or mode of establishment, and confusing them is the most common mistake.

Three statuses that the “findings” section distinguishes, and that ordinary reading conflates
Statut What it states Test that qualifies it Contresens habituel
Constatation An established, dated, sourced fact — including the absence of any anomaly on a system Reference to an item in the file or to a test Reading it as a criticism, when it may be neutral
Facteur contributif An element whose absence would have reduced the probability of the occurrence or mitigated its consequences Mental removal: does the sequence change? Ranking it by its position in the list, which is not a ranking
Cause An element without which the occurrence would not have happened as it did Same test, applied to the occurrence itself Confusing it with liability, which a safety investigation does not establish

The removal test explains why some elements, though visible, appear nowhere. A contributing factor is retained when its absence would have reduced the probability of the event or mitigated its consequences. A present condition but with no demonstrated effect on the sequence remains a finding: it is in the file, it does not weigh in the conclusions. This discipline gives the report its value and disappoints the reader looking for a single explanation.

Practice then varies from one organization to another: some distinguish an immediate cause from systemic factors, others publish a single, non-hierarchical list. Two reports of equivalent quality may therefore look very different; before comparing their conclusions, one must compare their drafting conventions, announced at the beginning of the section.

A recommendation is not an instruction, and its absence proves nothing.

A safety recommendation is addressed to a designated addressee: authority, manufacturer, air navigation service provider, operator. The latter is required to respond within a timeframe set by regulations, indicating the actions taken—not to apply it, and it can justify a refusal. A recommendation therefore never has, by itself, the scope of an airworthiness directive or operational requirement: confusing the two opens internal discussions without purpose.

Traceability chain, from the recommendation to the item in the fileA vertical arrow runs down four aligned levels: the recommendation, whose addressee is checked; the contributing factor, subjected to the removal test; the finding, established without judgement; and finally the item in the file on which everything rests. A level that does not lead to the next signals a reading to revisit.recommandationwho it is addressed tofacteur contributifremove it, the outcome changesconstatationestablished, without judgementitem in the filetrace, transcript, expert analysisa level without the next: reading to revisit
A recommendation that cannot be traced back to an item in the file is not usable in your organisation. Working back down the four levels from the recommendation is the quickest check of how sound a report is.

This does not mean that a recommendation addressed to a third party does not concern you: it identifies a defense deemed insufficient, and the useful question becomes “does this defense exist with us, in what form?” Conversely, the absence of a recommendation on a point analyzed at length does not close the subject: recommendations are not made for what has been corrected during the investigation, or outside its mandate. The paragraph on measures taken during the investigation then contains the answer.

Four common misconceptions in the preparation room.

Reading errors are not random: they are always the same, and each has a specific place to verify. Translation is added: a report has a reference version, the one published in the language of the investigation, and a vocabulary disagreement between two versions is decided on that one, where this vocabulary carries the status of the statements.

Four misreadings, what the report actually says and where in the document to check it
What people hear What the document establishes Where to look
“The report blames the crew” A crew action is a fact in the chain, just like a maintenance tolerance or an ambiguous instruction The objective paragraph, then the antecedents linked to that fact in the analysis
“The parameters show that…” A parameter has a sampling rate, a resolution and sometimes a computed origin; some traces are reconstructions The recordings appendix and the parameter list
“The cockpit voice recording says…” A transcript contains inaudible passages, uncertain attributions and a deliberate selection of sequences The conventions at the head of the CVR appendix
“No recommendation, so nothing to correct” The absence often signals a measure taken during the investigation, or a subject outside the remit The paragraph on measures taken during the investigation, and the appended comments

What the report becomes in your own documentation.

A report read is useless if it does not produce a written question. The transposition consists of three steps: isolating the findings that describe a configuration you know—same aircraft type, same airport, same subcontractor; identifying for each the corresponding defense in your approved documentation and verifying that it exists beyond the paper, i.e., who executes it and with what available time; recording the result in the safety management system, especially when it is negative, as this trace will allow the subject to be reopened at the next similar event.

Three men in an office review documents and discuss technical diagrams on a whiteboard.
Collaborative analysis prevents hasty judgments in investigations.

There remains a gray area. The follow-up of recommendations is rarely public in detail: it is known that a response has been provided, rarely what it produced two years later. And many instructive events close without a published report, failing to reach the threshold that triggers an independent investigation. The material closest to your operations is therefore in your own reports—the reports serving as a grid to interpret them.

This competence is noticeable. In selection interviews and online evaluations, describing the mechanism of a known event without moralizing, citing the place in the document where the information is established, immediately distinguishes those who have read the reports from those who have read only the summaries. This is what the training programs dedicated to flight safety and experience feedback work on, and what emerges from the exchanges on the professional forums when a report has just been published.

This article describes a reading method, not an obligation: the applicable texts, the official version of the concerned report, and the operator’s approved documentation prevail in all circumstances over what is written here.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

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