An NTSB investigation produces several kinds of public information over time. They do not all carry the same evidentiary meaning. Early material can establish that an event occurred and describe known conditions without determining probable cause.
Four layers to keep separate
First comes evidence collection: recorders, wreckage, weather, interviews, records, and other materials. Factual reports organize validated observations. Analysis connects facts and evaluates contributing conditions. The Board's probable-cause determination and safety recommendations are later outputs.
Recommendations are shown as a separate branch because preventing recurrence is not the same task as expressing a causal conclusion.
| Layer | What it can support | What it cannot yet support |
|---|---|---|
| On-scene evidence | observations, recovered material, recorded data | final causal conclusion |
| Factual record | verified sequence and documented conditions | Board determination by itself |
| Analysis | evaluation of relationships and contributing conditions | a conclusion before Board adoption |
| Probable cause | the Board's adopted determination | every possible contributing detail |
| Recommendation | a proposed safety action | proof that the recipient implemented it |
The exact documents and sequence vary by investigation. The table is a reading framework, not a promise that every case publishes the same artifacts on the same schedule.
The recurring reporting error
A preliminary report is sometimes presented as if it were the final explanation. The word “preliminary” does not mean unimportant; it means the document occupies an earlier stage in the evidence flow.
Another mistake is treating a factual statement as an analytical conclusion. “The recorder captured X” and “X caused the event” are different claims. The first can be a documented observation; the second requires analysis of timing, context, alternative explanations, and contributing conditions.
The opposite error also occurs: dismissing all early information because the final report is not available. A careful timeline can report confirmed event facts while leaving causal nodes explicitly unresolved.
When a source does not yet determine probable cause, the diagram should leave that node unfilled rather than using speculation to complete the story.
The evidence labels we use
Every node in a case-specific visual should carry one of four labels:
- reported: present in an official preliminary release or update;
- factual: supported by material in the public docket or factual reports;
- analytical: part of published analysis, with the document named;
- adopted: contained in a final Board determination or recommendation.
These labels describe source status, not our confidence score. A reported fact can later be corrected; an adopted finding can still have stated limitations. The source date and version remain necessary.
A case-specific timeline
For an individual investigation, replace the generic boxes with dated artifacts: preliminary report, docket release, hearing, abstract, final report, and recommendation updates. Each node should link to the official record and state whether it is factual, analytical, or final.
A useful timeline stores the event date separately from publication dates. Otherwise an article can make a new document look like a new incident development. The structure should include:
- event time and location;
- notification and investigative launch;
- preliminary public documents;
- docket evidence or hearings, if released;
- analysis and Board meeting;
- final determination and recommendations;
- later recommendation-response or closure updates.
Not every step will be public or applicable. Missing nodes should be labeled “not published” or omitted with a scope note, not guessed.
This is a better use of MyMap's timeline maker than a decorative sequence of news headlines because the stages explain what each document can support.
Questions the map should answer
A reader should be able to tell what is confirmed, what remains under investigation, which authority made each statement, and whether a safety recommendation has merely been issued or has a later response. If the visual cannot answer those questions, it is compressing the investigation rather than explaining it.
The map should not rank eyewitnesses, infer intent, simulate missing recorder data, or fill a causal gap from social-media footage. Those may be leads for reporters; they are not equivalent to findings in the official record.
Automation boundary
The NTSB's public data can surface new case records and changed fields. Automation can detect the update. It should not infer the importance or causal meaning of a change without reading the underlying record.
A production monitor should compare case identifiers, document lists, investigation status, report dates, and recommendation records. It should create an editorial review item—not automatically rewrite the causal diagram—when a new document appears. The editor then identifies which evidence layer changed and updates only those nodes.
This separation is the central quality control: machines detect that the record moved; a sourced analysis determines what the movement means.