An observation of nature needs explaining
A theory begins with something that does not yet have an accepted explanation — a pattern in the fossil record, an anomaly in planetary orbits, a repeatable effect in a lab. Science moves through a sequence, and each stage has to be completed before the next one is earned:
- Observation. Something regular or surprising is recorded — a measurement, a pattern across many samples, an effect that shows up again when someone looks for it. Once an observation has been confirmed repeatedly enough that no one seriously disputes it, it is treated as a scientific fact — not an explanation of anything, just a piece of the record that any later explanation has to account for.
- Hypothesis. A specific, testable explanation is proposed for that observation. It has to say what would happen under stated conditions, not just describe what was seen.
- Experiment and further observation. The hypothesis is tested directly, and independent researchers look for the same effect in new settings. Reviewers and replicators check that the reported result actually holds up, which is the same standard research-integrity bodies apply when they investigate whether a finding was properly obtained rather than fabricated or falsified.
- Accumulation of support. The explanation survives repeated, independent testing across many cases — not one experiment, but a body of converging evidence from different methods and different research groups.
- Acceptance as a theory. The scientific community treats the explanation as the best available account of the phenomena it covers, subject to revision if new evidence contradicts it.
The important point is that nothing skips a step. A single confirmed prediction does not make a hypothesis a theory; it takes an accumulated, cross-checked body of testing before the label is applied. That accumulation is also what quality-control efforts in scientific publishing are trying to track — flagging results that have not gone through the kind of independent replication described in recent work on automated screening of published findings. A claim that has only cleared step 2 is still a hypothesis, however plausible it sounds.
It helps to keep the three finished products of this process straight, because they answer different questions. A fact is a repeatedly confirmed observation — that species change over generations, that objects fall at a predictable rate. A law is a description of a regularity, often written as a mathematical relationship, that says what happens without saying why — the rate at which objects fall, stated as an equation, is a law. A theory is the explanation that accounts for the law and the facts together — why objects fall the way the law describes, and why that same underlying mechanism also explains orbits, tides, and the paths of satellites. Laws describe, theories explain, and facts are the observations both have to be consistent with.
People hear 'theory' in its everyday sense of a hunch
In ordinary conversation, "theory" often just means a guess — "I have a theory about why the bus is late." That casual usage is not wrong for everyday speech, but it is a different word from the technical one, and treating them as interchangeable is where most public arguments about science go wrong.
The confusion runs in one direction: people hear "theory" used for an unsupported hunch, and conclude that a scientific theory must likewise be unproven. It isn't. The distinction is fixed by two things, not by tone or confidence:
| Hypothesis | Scientific theory | |
|---|---|---|
| Scope | One specific, testable claim | Explains a wide range of related facts and observations |
| Support | Untested or minimally tested | Repeatedly tested and confirmed across independent lines of evidence |
| Status | Provisional starting point | Best-supported explanation currently available |
| Revisability | Discarded easily if the first test fails | Modified or narrowed, rarely discarded outright |
Plate tectonics and evolution are the standard examples precisely because they show what "theory" buys you at that level: not a single experiment but decades of independent evidence — seafloor mapping, seismology, genetics, the fossil record — that all converge on the same explanation. Calling either one "just a theory" is treating step 5 of the pipeline above as though it were step 2. The everyday sense of the word describes an untested idea; the scientific sense describes an idea that has already survived the testing that would have killed it if it were wrong.
New observations conflict with an accepted theory
A theory that has been broadly accepted for years can still run into observations it does not fit. That is not a scandal — it is the normal condition under which theories get better.
The sequence when this happens looks like this:
- A new observation or measurement produces a result the current theory did not predict.
- Other researchers try to reproduce the anomaly, because a single unreplicated result could be measurement error, contamination, or a mistake in analysis rather than a real conflict. This is the same replication-and-review filter that scientific quality-control work is built around — separating a genuine anomaly from noise in the data pipeline.
- If the anomaly holds up under independent scrutiny, the theory is adjusted: its scope may be narrowed (it still applies, but only under certain conditions), it may be modified to account for the new case, or — if the conflict is deep enough — it is eventually replaced by a better-supported explanation.
- Whatever replaces it still has to explain everything the old theory explained. A successor theory doesn't get to drop the facts that were already accounted for; it has to cover the old ground and the new anomaly.
That last constraint is why a superseded theory was rarely simply "wrong." Newtonian mechanics still predicts the motion of thrown balls and orbiting satellites correctly within its working range; it was narrowed, not discarded, once relativity extended the picture to conditions Newton's framework didn't cover. The law describing how a falling object accelerates did not change when the theory behind it was extended — the regularity itself still held, only the explanation for why it holds got deeper. Revisability is what keeps a theory scientific — an explanation that could never be modified in light of new evidence would have stopped being science and become dogma.
A claim is examined for whether it predicts anything observable
Before any of the above matters, a claim has to pass one test: could an observation, in principle, contradict it? That single question is what separates science from everything that merely sounds scientific.
- If no possible observation could count against the claim — it's compatible with any outcome — it is untestable. It can be discussed, believed, or debated, but it stays in the category of opinion, not science.
- If some possible observation could contradict the claim, the claim is falsifiable. It can then be tested directly, refined into a specific hypothesis, and — if it survives repeated, independent testing across many cases — eventually earn theory status.
This is the criterion that quietly runs underneath every stage described above, even though it rarely gets stated as a rule on its own: testability isn't one property among several that a good scientific claim happens to have, it's the gate a claim has to pass through before observation, experiment, or replication can do anything with it at all. A claim that predicts nothing checkable can't be pushed toward hypothesis or theory no matter how much evidence accumulates around it, because there's nothing for the evidence to confirm or contradict.
Applying that one question — what observation would prove this wrong? — to any claim encountered next, whether it's framed as a scientific finding or simply asserted as fact, is the fastest way to tell which category it actually belongs in.