The Stopping Rule is an interactive HTML demonstrator designed to explore how clinicians, medical students and clinical trainees make decisions under uncertainty, with particular emphasis on when they decide that they have enough evidence to stop considering alternative explanations.
The simulator is based on problem-based learning principles. In PBL, the tutor does not simply provide the answer. The tutor acts as a metacognitive questioner, challenging assumptions, asking what evidence supports a conclusion, identifying what does not fit, and encouraging learners to consider what additional information would materially change their judgement. The simulator applies this same approach to professional and clinical reasoning.
Two cases are currently included: Dr Birt and Dr Robin Smith. Both involve questions of competence, confidence, evidence and alternative explanations. They were selected because they allow reasoning to evolve as additional information is disclosed, rather than presenting the participant with a complete case from the outset.
Each case is presented sequentially. At every stage, the participant is asked to record:
- their current judgement;
- their confidence in that judgement on a 0–100% scale;
- whether they would keep the problem open, close it, reopen it, or remain uncertain;
- what information would be most likely to change their judgement;
- what evidence or features of the case currently do not fit their explanation.
The purpose is not simply to determine whether the participant reaches the “correct” conclusion. Instead, the simulator records the trajectory of reasoning: how confidence changes, when closure occurs, whether contradictory evidence is recognised, and whether a previously closed judgement is reopened when new information appears.
The simulator contains two experimental conditions.
In Condition A, participants work through the case without additional assistance. This provides an unaided reasoning baseline.
In Condition B, participants encounter the same case with metacognitive prompts. These prompts do not tell participants what conclusion to reach. Instead, they challenge the reasoning process. For example, the simulator may ask whether a judgement is based on direct evidence or inherited opinion, whether different competence domains are being conflated, what evidence would falsify the current interpretation, or whether alternative explanations have been adequately considered.
At the end of the case, the simulator compares the two reasoning trajectories. It shows the stage at which explicit closure occurred, whether the case was subsequently reopened, changes in confidence, identification of contradictory evidence, and the extent to which participants specified information that could materially alter their judgement.
The underlying educational and research question is whether metacognitive intervention can improve the quality of the stopping rule.
This is important because poor decisions often arise not from lack of knowledge, but from closing a problem too early. A practitioner may reach a plausible conclusion and then progressively interpret subsequent evidence through that initial framing. The simulator therefore treats closure as a decision under uncertainty rather than simply as a confidence threshold.
The present version is a demonstrator rather than a validated assessment instrument. Its purpose is to make reasoning visible and experimentally tractable. A future clinical version could use evolving diagnostic cases to investigate whether metacognitive support reduces harmful premature diagnostic closure without encouraging unnecessary investigation or perpetual uncertainty.
The broader proposition is that AI may add clinical value not simply by producing better answers, but by helping clinicians judge when their current answer is not yet sufficient to justify stopping.