Human error types and troubleshooting: from slips, lapses and mistakes to recovery information

What the research shows

Three levels of behaviour. Jens Rasmussen (1983) described human performance at three levels. At the skill-based level, routine actions run with little conscious control. At the rule-based level, people recognise a situation and apply a stored rule ("if the pressure light is on, open the bleed valve"). At the knowledge-based level, in situations no rule covers, they reason from their understanding of the system, slowly and with effort.

Error types. James Reason (1990) built on this framework. Slips are actions carried out differently from the plan, often when attention is elsewhere: the right intention, the wrong lever. Lapses are memory failures: a step is omitted, a place in the procedure is lost after an interruption. Mistakes are failures of the plan itself: a good rule is applied in the wrong situation, a bad rule is applied, or reasoning at the knowledge-based level goes wrong because the person's mental model is incomplete. Violations, deliberate deviations from rules, are a separate category with different causes. The theory glossary lists skill-, rule- and knowledge-based behaviour and slips, lapses and mistakes as terms of their own; Donald Norman's treatment of error in The Design of Everyday Things uses a similar distinction.

Awareness. Mica Endsley's (1995) model of situation awareness adds that many errors start with a wrong picture of the situation: something was not perceived, not understood or not projected correctly.

What it means for documentation

• Organise troubleshooting by what the reader observes. A troubleshooting topic starts from the symptom, the message or the visible state, then gives the possible cause and the remedy. This supports the rule-based level directly; the DITA troubleshooting topic type follows the same pattern.
• Give a model for the unexpected. When no symptom table fits, the reader works at the knowledge-based level. A short explanation of how the system works, its states and what each indicator means, lets the reader reason instead of guess.
• Support memory where lapses occur. Checklists, numbered steps, visible progress and a clear restart point after interruptions reduce omitted steps.
• Accept that slips happen. Slips are hardly prevented by text. They call for design: interlocks, confirmations, consistent controls and well-designed alarms (see alarm management). Documentation can say how to detect a slip and what to do next.
• Write recovery information. Say how to recognise that something went wrong, how to bring the machine or process into a safe state, how to undo the action if possible, and when to stop and call a qualified person.

Limits

The taxonomies are frameworks for analysis, not measurements; real events often mix types. In machine safety, human error and reasonably foreseeable misuse are part of the risk assessment, and protective measures follow the order of inherently safe design, safeguarding and only then information for use. No cognitive argument replaces the risk assessment or legally required safety information: troubleshooting text never stands in for a design measure, never blames the user for a foreseeable error and never makes a product compliant.