In one line: do not multiply entities beyond necessity — when two explanations fit the same evidence, consider the one with fewer assumptions first.
The name refers to the 14th-century English logician William of Ockham. The Latin tag often quoted — "entities should not be multiplied beyond necessity" — is a later summary rather than his own phrasing; his wording was closer to "a plurality is not to be posited without necessity". The core claim is the same: do not add unnecessary assumptions to an explanation.
It is a methodological principle, not a law of nature, and not "the simplest answer is always correct". Its job is to give you a sensible order in which to consider options when evidence is thin.
Suppose two explanations account for what you observed:
The razor suggests considering A first. Not because A is simpler and therefore "true", but because B introduces more unverified links, each of which can fail — and their probabilities multiply down.
Note the word "first". If new evidence supports B, you move to B. The razor orders options; it does not convict.
Rhythmic knocking at night. "Something mysterious" needs an extra assumption; "a timed appliance or pipe expansion" needs one, and such things are common. Check that first.
A slow laptop. Suspect too many background apps or a nearly full disk before a hardware fault — the former is testable in a minute.
A colleague not replying. "He is annoyed with me" needs several assumptions (he noticed, he cares, he chose to ignore); "he is busy today" needs one.
A software error. Misconfiguration, permissions and dependency versions are far more common than a bug in the underlying framework.
Choosing between scientific models. Between two models that fit the data equally well, the one with fewer parameters and better falsifiability is preferred.
Misuse 1: treating "simpler" as "correct". The razor ranks options; it does not decide. Rare explanations do sometimes turn out to be true, especially with strong evidence.
Misuse 2: using it to dismiss low-probability options. "Unlikely" is not "impossible". The correct statement is: "with current evidence it is low probability, so set it aside — but revisit it if this kind of evidence appears."
Misuse 3: ignoring base rates and evidence strength. Judge probability not only by how vivid a story is, but by how common the thing is (the prior) and how strong the new evidence is (the likelihood). See Bayesian thinking.
Write the observations as clues, enumerate every plausible cause, rank by number of assumptions plus base rate, then design one verification action for the top option. That is exactly how Ockham works — try it on the bench, or read the method.