From a task to a sequence of events

THERP was developed to support probabilistic safety assessments, particularly for nuclear installations. It represents tasks through more detailed actions and examines how their possible outcomes contribute to successful or failed performance of a function.

Although it can be integrated with fault tree analysis, the two should not be confused. Human reliability event trees represent sequences, including opportunities to detect and recover from an error before the consequence under consideration occurs.

Operationally meaningful detail

Decomposition should make clear what the person does, with which information and under what conditions. Separating every movement without analytical purpose can create a huge model that neither explains risk better nor helps improve the task.

Preparation combines documentary review with input from operating personnel. For example, a procedural check may not be independent of the preceding action if it uses exactly the same indicator or relies on the same interpretation of an ambiguous condition.

Dependence and recovery

The method addresses error probabilities and factors affecting performance, but treatment of dependence is essential. Multiplying probabilities as though every action were independent can underestimate the risk associated with the complete sequence being assessed.

Recovery opportunities need evidence. There must be a real chance to detect the problem, understandable information, sufficient time and the ability to act. Adding a second person to the diagram does not by itself demonstrate that an error will be corrected.

Deliverables and limitations

A traceable assessment documents tasks, assumptions, data sources, operating conditions and sensitivity of results. It also identifies practical measures such as interface, verification or planning changes and explains which assumptions would need reviewing after their implementation.

THERP requires specific training and can be resource intensive. It is not a universal calculator for assigning reliability to employees. Its value depends on whether the model represents the operation and keeps uncertainty visible, especially when data are transferred between sectors with different tasks, interfaces and operating conditions. These differences should remain explicit in the assessment.