You take a remedy and feel better the next day. The sequence is real; the causal conclusion is not established. Most minor illnesses resolve on their own, and you would very likely have improved anyway — which means the improvement following the remedy is exactly what you would expect whether or not the remedy did anything at all.
Temporal order is a genuine requirement for causation: causes precede effects. But it is a filter, not a proof, and treating it as proof is one of the costliest errors in ordinary reasoning. At least three other explanations always compete. It may be coincidence — with enough events, striking sequences are guaranteed, and memory obligingly preserves the hits while discarding the misses. It may be reverse causation — B produced A, and the sequence you noticed was misleading. Or there may be a common cause producing both, which is the possibility people miss most often and the reason ice cream sales track drowning deaths.
The medical version is the most consequential, and it is why controlled trials exist. Regression to the mean guarantees that people typically seek treatment when symptoms peak, so improvement afterwards is the statistical expectation regardless of what was taken. Add the placebo effect and the natural course of illness, and post hoc reasoning will validate almost any intervention. The entire apparatus of randomisation and control groups exists to defeat precisely this fallacy — which is a good measure of how powerful it is.
Testing a causal claim means asking what else could explain the sequence. Does it hold up when repeated? Does the effect scale with the dose? Is there a plausible mechanism? Does it survive when you control for other factors? None of these is individually decisive, but together they are the difference between noticing a sequence and establishing a cause.