The Selection Window: Why "Finish the Course" Had the Physics Backwards

For eighty years, "finish the full course" functioned less like medical advice and more like a moral test. Stop early and you're careless, maybe complicit in the coming superbug apocalypse. You've privately judged the friend with the half-full amoxicillin bottle in the medicine cabinet.

Here's the uncomfortable part: your friend was right, and the guilt you've been carrying runs on a mechanism pointed in exactly the wrong direction.

The old story: bacteria are an army, antibiotics are an assault, and if you stop firing before the last soldier is dead, the survivors — the toughest, most resistant ones — regroup and come back stronger. Finish the course. Leave no stragglers to breed a resistant strain.

It's a great story. A 2023 review of more than 120 randomized controlled trials spanning 25 years found it "rather based on urban legend." Short courses work just as well as long ones for conditions from urinary tract infections to pneumonia — and the resistance risk runs backwards from the folklore. Longer exposure to an antibiotic breeds more resistance, not less.

Why, exactly. Resistance isn't attrition — it's selection. Every day a bacterial population sits inside a body bathed in a drug, only the bacteria that can tolerate that drug survive to reproduce. The drug doesn't just kill susceptible cells; it curates the resistant ones, generation after generation, for as long as the exposure continues. Push the course past the point of clinical need and you're not mopping up stragglers — you're running an extended breeding program for the exact trait you're trying to avoid. Call this the Selection Window: the period during which a population is exposed to a pressure it can adapt to is also, mechanically, the period during which it evolves resistance to that pressure. The window itself is the risk. Shrinking it — not intensifying what happens inside it — is what limits resistance.

This is worth carrying as a general model, because it applies wherever a pressure is sustained against a population capable of adapting to it: pesticides on insect populations, chemotherapy on tumor cell lines, rate-limiting attempts against a system that logs and adjusts to failed logins. The instinct is almost always "apply more force, for longer, to be thorough." But when the target can evolve in response to the pressure itself, duration of exposure — not intensity of the assault — is often the variable doing the damage. So the diagnostic question changes. Not "did I go hard enough, long enough?" but "how long am I giving this thing a chance to learn from the pressure I'm applying?" Sometimes the responsible move is to hit hard and stop — grinding on is the mistake, not the caution.

The second thing worth understanding here is how bad advice survives eighty years inside a field built on evidence. It wasn't stupidity. "Finish your course" is simple, memorable, and moralized — which made it excellent public health messaging and an almost un-revisable piece of doctrine once entrenched. A recent survey of 1,400 adults found 90% still believe you should always finish the full course, and 60% actively prefer longer treatment, on the reasoning that more pills must mean more safety. That reasoning isn't stupid in general — "better safe than sorry" is a sound heuristic almost everywhere. It fails here specifically because the harm you're being protected from is manufactured by the safety measure itself. Same structural trap behind decades of blanket dietary rules, or "patch every system to the maximum, always," or any policy that got adopted because it was easy to state and became politically expensive to walk back without the institution appearing to admit it was wrong for decades.

None of this is an argument for distrusting doctors — it's an argument for building tolerance for advice that updates, because updating in the face of new evidence is the system working, not failing. Here is what to actually do with it. Next prescription, ask your doctor two specific questions: whether a shorter, evidence-based course applies to what you have (guidelines for pneumonia, UTIs, and several other common infections now specify durations far shorter than the old five-to-fourteen-day defaults), and whether you should stop based on symptom resolution rather than a fixed calendar count. Don't self-truncate without asking — the point isn't "trust yourself over medicine," it's "the guideline has probably already changed, go find out." And more generally, next time you're told to apply a fixed, generous duration of any pressure "to be safe" — a security policy, a training regimen, a punishment — ask whether the thing on the receiving end can adapt to that pressure over time. If it can, the calendar you were told to follow is doing more of the damage than the intensity ever was.

And maybe extend an apology to that friend.

Distilled from Nautilus