Part 1 · Healthcare & Patient Safety · What failed

Therac-25

About this resource · How to use it · Learning engineering connections

What is this resource?

A documented case about Therac-25. Six known massive radiation overdoses. Three patients died from complications attributed to the overdoses.

How should I use it?

  1. Read the narrative and five beats, keeping the disclosures and scope limits in view.
  2. Compare the LE insight with the LENS approach. Identify the system condition that changed what people could do, then check the cited evidence before applying the lesson elsewhere.

How does this connect to LENS and learning engineering?

A therapist’s ability to detect and respond to danger depended on truthful feedback, a usable stop and reliable incident escalation. The machine also needed an independent safeguard that worked within the hazard’s timescale. These are separate requirements. Use the Tyler record to propose and test a therapist’s ability to interpret a discrepant reading, choose a safe stop and report the incident. Vary cryptic codes and monitor failures in a simulator and assess unfamiliar faults. Test independent beam inhibition separately. The historical record cannot show whether that proposed practice would have changed the outcome.

1985 – 1987 healthcaretech

Impact. Six known massive radiation overdoses. Three patients died from complications attributed to the overdoses.

Potential learning engineering response. The record identifies a capability gap that learning engineering could directly address. The proposed response still needs testing.

Ask AI about this case Opens a new chat prefilled with this case — explore it in your own context.
Consider the human, the science and the system
Human
Who gained or lost agency, opportunity or well-being? Whose experience is missing from the record?
Science
What evidence supports the proposed learning explanation, and what other causes could explain the outcome?
System
Which tools, responsibilities, interfaces or conditions enabled the result or prevented recovery?
Explore the three foundations →

A therapist correcting a prescription quickly exposed a software defect. The console then showed an apparent underdose and allowed a retry, even though the patient had already received a massive overdose. The Tyler record gives us a specific question to test in simulation: what information and authority would let a therapist recognize a dangerous discrepancy, stop treatment and report it? A separate test would verify that the machine independently prevents an unsafe exposure.

In brief

Therac-25’s design did not carry forward some of the independent hardware safeguards used in earlier machines. Leveson and Turner describe two established software defects: a fast-edit race condition in two Tyler events and a counter overflow in the fatal January 1987 Yakima event. In Tyler, the console displayed an apparent underdose, the manual did not explain the fault code, and the system allowed a one-key retry. Three of the six accidents lack definitive technical explanations. The corrective action plan combined hardware, interface and reporting changes.

The case in five beats

  1. Earlier independent safeguards were only partly carried forward
  2. Two Tyler events involved a fast-edit race condition
  3. A distinct counter overflow caused the fatal January 1987 Yakima event
  4. In Tyler, the interface showed an apparent underdose and permitted a retry
  5. Hardware, interface and reporting changes arrived together
The Learning Engineering Lens

LE insight

A therapist’s ability to detect and respond to danger depended on truthful feedback, a usable stop and reliable incident escalation. The machine also needed an independent safeguard that worked within the hazard’s timescale. These are separate requirements.

LENS approach

Use the Tyler record to propose and test a therapist’s ability to interpret a discrepant reading, choose a safe stop and report the incident. Vary cryptic codes and monitor failures in a simulator and assess unfamiliar faults. Test independent beam inhibition separately. The historical record cannot show whether that proposed practice would have changed the outcome.

The LENS competency this case exercises
  1. 1 Systems Analysis
  2. 2 Iterative Development
  3. 3 Human-System Collaboration
  4. 4 Test & Evaluation
  5. 5 Sociotechnical Constraints

Problem type · PT3

Problem type
D3/PT3
Induced
3.1
LEO
3

Competing readings

Compare what the sources agree on, where their findings differ, and how each interpretation would change what you test or design.

  • Independent interlocks could prevent a hazardous exposure. Truthful monitoring and an appropriate stop could have prevented a repeat after the first exposure. The corrective action plan changed both.
  • Another reading emphasizes reporting and escalation. Connecting earlier incidents across sites might have changed the response, but the available record cannot establish the effect of that counterfactual.

What this case does not show

The six cases are a documented lower bound. Three incidents lack definitive technical explanations, and the corrective action plan changed several safeguards together.

Sources and evidence