A diagnostic test is most informative when it is applied against a strong prior intuition, not scattered indiscriminately. Ordering a test when you already have a confident hunch sharpens or overturns that hunch; ordering tests blindly on low-suspicion cases generates false positives that mislead more than they reveal.
The exceptional case that defies expectation is not noise to be discarded but the most valuable data point, because it exposes a rule the normal cases obscured. Studying why one patient responds when the model says they should not can unlock the mechanism governing all the others.
Medicine is reframed not as the application of fixed biological laws but as reasoning under irreducible uncertainty, where every decision is a probability weighed against incomplete information. The practitioner's skill lies in navigating what cannot be known precisely rather than in eliminating doubt.
No clinical observation is neutral: the observer's prior beliefs shape what data gets collected, noticed, and recorded, introducing systematic bias into supposedly objective measurement. Recognizing this bias is itself a corrective discipline, because the normal human instinct is to see confirmation of what one already expects.
Mukherjee recounts the eighteenth-century minister Thomas Bayes, who asked how a prior belief should be revised in light of new evidence, formalizing how to reason backward from observed effects to their likely causes. This becomes the template for how a doctor should update a diagnosis as test results arrive rather than treating each result in isolation.