Build a causal account from evidence

An investigation reconstructs an event using telemetry, physical evidence, test results, procedures and accounts from people involved. Each source has limits: a sensor may stop recording, hardware may be unavailable and memory may be incomplete. A useful timeline separates observed events from inferred explanations.

A plausible story is not enough. Investigators compare candidate mechanisms with the evidence and test whether an explanation accounts for the sequence. When evidence cannot uniquely identify a cause, the report should retain a most-likely or unresolved classification rather than convert confidence into certainty.

A direct mechanism — open definition might be a leak, a broken connection or an incorrect command. Contributing conditions can include design assumptions, testing gaps, unclear responsibilities or information that did not reach a decision-maker. Examining these layers helps explain why the event occurred and which barriers failed to prevent it.

NASA and NOAA’s GOES-17 investigation, for example, examined an instrument cooling problem and identified a most likely cause with associated contributing factors. The wording matters: an investigation’s probability judgment should not be rewritten as if every internal event was directly observed.

In a fictional test, a sensor connector loosens after vibration and no acceptance check detects it. Reattaching the connector addresses the immediate condition, but the investigation should also ask about retention design, representative testing and inspection. Different findings can require different corrective actions.

A good account does not treat a human mistake as the end of explanation or assume that systems analysis removes individual responsibility. It asks what actions occurred, what information and authority people had, and how the organization shaped their choices. The purpose is an evidence-based explanation and effective prevention, with respect for the people affected.