How Aviation Safety Teams Separate Fact From Hypothesis Today

Aviation safety management rests on a plain observation about human behaviour: when something unexpected happens, the first account of it is almost always a blend. Part of it is what was actually seen or recorded. Part is what the person present assumed. Part is a theory about why it happened. The practical question for any safety team is how to keep those parts distinguishable long enough to act wisely. This piece looks at how current practice approaches that problem, without claiming any single method is universal.

It is an overview written from general, public principles of safety management. It is not operational guidance, and specific procedures always come from the relevant regulator, the operator’s manuals and approved training.

The problem: first reports are blended

Anyone who has read a raw event report will recognise the pattern. A short narrative says that a component “failed”, the crew “lost” an indication, a procedure “was not followed”. Each of these phrases bundles an observation with an interpretation. “The indication disappeared” is something a person saw. “The sensor failed” is a conclusion about cause. If both sit in the same sentence, later readers inherit the conclusion as if it were data.

Good reporting systems therefore encourage the reporter to describe what happened in plain, observable terms first, and to place interpretation in a separate field or paragraph. The structure itself nudges people toward honesty about the edge of their knowledge.

A common three-layer structure

Although terminology varies between organisations, many teams end up with three layers in an investigation record.

  • Established facts: things supported by recorded data, physical evidence or direct, documented observation, each with its source and time.
  • Working hypotheses: candidate explanations that fit the facts so far, each listed with what would support or undermine it.
  • Open questions: things nobody yet knows, along with who is expected to find out and by when.

The value is in the separation, and in how entries move between layers. A hypothesis can graduate into an established fact only when something supports it. A fact can be demoted when its source turns out to be weaker than assumed. The record stays alive and honest instead of hardening around the first plausible story.

Why hypotheses need an owner and a test

A hypothesis without a test is just an opinion that is hard to challenge. Mature teams attach a next step to each one: a data pull, an inspection, an interview, a simulation. That turns a debate into a task. It also reduces the pull of the most senior or most confident voice in the room, because the question becomes “what would show this is wrong?” rather than “who agrees?”

One benefit that practitioners often mention is that it makes disagreement cheaper. Two people can hold different hypotheses without conflict, provided both are listed and both have a test pending.

Data volume changes the stakes

Aircraft and operations now generate far larger streams of recorded data than they did a few decades ago, and more of it is analysed automatically. That is generally a benefit, because it expands the set of things teams can confirm rather than infer. It also creates a new risk: output from an analysis tool can look authoritative simply because it is precise. A trend line is a measurement of the data supplied to it, not a verdict on cause. Teams that handle this well label tool outputs as observations about the data and keep interpretation in the hypothesis layer.

Culture matters as much as format

A template cannot do the work alone. If people fear that saying “I assumed” will be punished, they will hide assumptions inside confident language, and the structure becomes decoration. Organisations that report better results from their safety processes tend to describe an environment where acknowledging uncertainty is treated as a professional strength. The language of a just culture, where honest reporting is protected, is largely about making that possible.

Managers in these settings model the behaviour. When a manager says “I do not know yet, and here is how we will find out”, it sets a standard that is easy for others to follow.

Typical weaknesses seen in practice

Several recurring problems appear in discussions of investigation quality, and each is a failure to keep the layers separate.

  • Premature closure: settling on the first coherent explanation and treating it as established.
  • Source drift: a fact quoted repeatedly until nobody remembers who first reported it or when.
  • Hindsight blending: describing what people “should have known” using information that only became available afterwards.
  • Orphan questions: open items with no owner, which quietly disappear.

A useful periodic exercise is to take a closed report and re-sort its sentences into the three layers. Many teams find that some conclusions rested on less than they appeared to.

Where the same pattern appears elsewhere

The layered record is not unique to aviation. Software and infrastructure teams use something very close when they respond to an outage: a timeline of observed events, a list of competing theories under test, and a list of unanswered questions. Engineers interested in that world might read around on the DevOpsSchool site, which pairs self-paced lessons and hands-on labs with quizzes and cheatsheets for software engineers, where responders’ habits are taught as part of broader technical training. The overlap between the two fields is a reminder that the discipline is about information, not about any particular machine.

Training and handover

The habit also has to be taught. Training programmes increasingly use scenario debriefs in which participants are asked to restate an event in three passes: first only what was observed, then what they inferred, then what they would test next. Handovers between shifts or departments benefit in the same way, because the incoming person receives not just a story but a map of how firm each part of it is.

What to watch for next

Looking ahead, the general direction is clear enough. More data, more automated analysis, and more attention to the human side of reporting all point the same way: toward records that make the status of each statement explicit. Expect continued emphasis on traceability of sources, on time-stamping information, and on treating uncertainty as something to be managed openly rather than smoothed over.

A practical takeaway for any team

You do not need a formal programme to adopt the core idea. For your next review of any incident, small or large, rule three headings on a page: Established, Hypotheses, Open. Move each statement under the right one, add a source and time to every established fact, and attach a test and an owner to every hypothesis. Revisit the page whenever new information arrives. The structure is simple, but it keeps a team honest about what it knows at any given moment, which is the real purpose of the whole exercise.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top