Flight data monitoring turns recorded information into questions that a safety or engineering team can investigate. Its usefulness depends on reliable data, clearly defined events, appropriate context and follow-through. Start by checking what the records represent before interpreting a count or trend. The fictional exercise below demonstrates that discipline without supplying aircraft operating limits or a production monitoring algorithm.
On a narrow screen, scroll tables sideways. Keyboard users can focus a table and use the arrow keys.
Understand the monitoring loop
The FAA lists Advisory Circular 120-82, issued in 2004, as active guidance for a voluntary Flight Operational Quality Assurance (FOQA) program. The circular addresses data validation, event validation, corrective action and follow-up. In Europe, EASA Decision 2025/020/R updates guidance on FDM implementation; its page identifies corrected documents issued in March 2026.
A useful conceptual loop is collect, validate, review, act and evaluate. First establish whether the recording and conversion are trustworthy. Then review a flagged event in context. If action is warranted, assign an owner and a way to assess the result. Organizational approvals, confidentiality and reporting arrangements belong to the applicable program.
Check data quality before counting events
Describe the dataset before plotting it: aircraft or fleet, recording period, available parameters, units, sample rates, conversion version and missing-data handling. Keep received files separate from unique flights and usable observations. The count of files delivered can be larger or smaller than the population suitable for a particular analysis.
| Check | Example issue | Effect on the analysis |
|---|---|---|
| Identity | The same flight arrives twice | A duplicate can inflate both flights and events. |
| Completeness | A required interval is missing | The flight may be unusable for this question. |
| Meaning | A field changes unit or definition | Values become incomparable without documented conversion. |
| Timing | Sources have different clock references | Apparent event order may be misleading. |
| Context | A flag has an operational explanation | A candidate event needs review before a conclusion. |
Use a quality rule that fits the question and document it before interpreting the results. Preserve the reason for each exclusion. If a recording change affects one fleet or period, examine whether an apparent trend reflects the measurement process.
Work through a fictional denominator example
Download the fictional monitoring sample (JSON) and open it in a text editor or a JSON-capable analysis tool. It contains 12 received summary records, not raw flight-recorder data. Fields define a synthetic 60-sample observation window, the number present and a preassigned review flag. The flag has no aircraft-safety meaning.
- Count the received records: 12.
- Deduplicate by
flight_id. F07 appears twice with identical values, leaving 11 unique flights. - For this exercise only, retain flights with all 60 expected samples. F11 has 42 and is excluded, leaving 10 usable flights.
- Count flagged usable flights. F03 and F08 are flagged, giving 2 ÷ 10 × 100 = 20%.
Dividing the same two flags by all 12 received records gives 16.7% when rounded to one decimal place. That denominator includes a duplicate and an incomplete record. The exercise demonstrates why the population definition belongs beside the percentage.
The printable data-review sheet records these counts and the exclusion reasons. The JSON also includes field definitions, assumptions and expected results so the exercise remains understandable when saved separately.
Turn a flag into a reviewable question
A candidate event identifies records for review. The FAA circular explicitly distinguishes data validation from deciding whether an apparent event is valid in its operational context. A qualified program defines its own event criteria, checks the underlying measurements and considers relevant reports and circumstances.
For a real analysis, report both the rate and the counts, with the period, eligibility rule and data-quality exclusions. Compare like populations. Ten flights are enough to demonstrate arithmetic, but this synthetic sample provides no evidence about an airline, crew or safety trend.
Keep authorized data access and retention rules attached to the workflow. Before using an external analytical service, confirm that the organization permits that data and purpose. After a reviewed finding leads to action, record who owns it, the expected effect and when comparable data will be reviewed again. That closes the loop between observation and improvement.