Verification

Verification is the demonstration or confirmation that a statement is true. For particular statements it is achievable. For general ones it is the problem this whole page is about.

A hypothesis is a supposed statement whose correctness is to be tested empirically. It is formulated so that a test could go either way. A supposition no test could touch is not a hypothesis.

Falsification is the refutation of a hypothesis by a contrary observation. It is logically decisive where verification is not. One counter-instance settles a general claim; a million instances do not.

Critical rationalism holds that general statements can only ever be refuted, never established. It replaces the search for proof with the search for error. It is the position that followed the failure of verificationism.

Karl Popper argued that verification of general statements is impossible and that falsification is the workable criterion. He shifted the question from how to confirm to how to expose. The shift is the origin of modern scientific method as it is taught.

Logical empiricism held that the correctness of hypotheses could be established by observation. It was the programme Popper opposed. Its collapse is the reason the rest of this page exists.

Positivism assumes that empirical claims admit of verification. It is the wider tradition behind logical empiricism. Its appeal is that it makes knowledge look secure.

Verificationism defines the meaning of a sentence by the method of its verification. A sentence with no verification procedure is not false but meaningless. The criterion turned out to be unable to satisfy itself.

The Vienna Circle used verificationism to separate scientific statements from metaphysics. It was the most consequential philosophical programme of its century and it failed on its own terms. What survives is the demand that claims be testable.

The problem of induction shows that no number of particular cases logically warrants a general statement. It is the reason verification cannot work for laws. Every practical method since is a response to it.

A criterion of meaning separates scientific statements from untestable metaphysics. It answers which sentences are candidates for truth at all. The attempt to state one precisely is what broke the programme.

Corroboration describes how well a hypothesis has survived severe tests. It is deliberately not a measure of probable truth. A well-corroborated theory is one that has been attacked and has held, so far.

Rudolf Carnap developed successive versions of verificationism, weakening the criterion each time to save it. The sequence is instructive: each repair narrowed what the criterion could do. He is the record of an honest retreat.

A demarcation criterion marks the line between empirical science and what lies outside it. Popper proposed falsifiability. The line matters because it decides what has to answer to evidence.

Validation checks whether something meets its intended requirements, which is a technical rather than an epistemological question. Building the thing right and building the right thing are different checks.

It stays with the historical controversy between verificationism and falsificationism. It leaves out everything that replaced both in practice: confirmation theory, probabilistic updating, the statistical machinery of testing, and the contemporary evidence that the machinery is not working as intended.

A protocol sentence is the record of an observation, taken as the base on which theory rests. The debate about whether such sentences are theory-free was the central dispute of the Vienna Circle.

Confirmation theory asks how and how much evidence supports a hypothesis, rather than whether it proves it. The question moved on.

The Duhem-Quine thesis holds that hypotheses are tested only together with auxiliary assumptions. When a test fails, the thesis says the test cannot tell you what failed.

Bayesian confirmation updates a degree of belief in the light of evidence, by a stated rule. It is the formal successor to the whole dispute.

Hempel's paradox shows that a green leaf confirms the statement that all ravens are black, if confirmation works as stated. It exposes a real defect in naive confirmation.

Otto Neurath opposed the idea of theory-free protocol sentences within the Vienna Circle itself. Beside Carnap he is the other side of the central argument.

Falsifiability is the property of a statement that some possible observation would refute it. The property is what can be assessed before any test is run.

Construct validity asks whether an instrument measures the concept it claims to measure. It is the methodological form of the verification question.

Statistical hypothesis testing is how hypotheses are actually examined, using samples and error probabilities. It is the practical descendant of the whole dispute.

A type I error is rejecting a true hypothesis. It quantifies the risk that a refutation is itself mistaken.

Operationalisation is the translation of an abstract concept into observable indicators. It is the step at which a hypothesis becomes testable at all.

Alfred Tarski gave a formal theory of truth separating object language from metalanguage. It is the technical resolution of problems the Vienna Circle wrestled with informally.

Model checking is automated verification of whether a system satisfies a specification. It is the one domain where verification in the strict sense actually succeeds. Applied logic solved a version of the problem philosophy abandoned.

The replication crisis is the contemporary finding that a large share of published results do not reproduce. It is the empirical verdict on how testing is practised.

If a significant result verified its claim, repeating the study would return the same verdict. About a third of repetitions did.

Triangulation secures a finding by combining independent methods. It is the practical substitute for proof. Agreement between methods that fail differently is the closest thing to verification available.

Particular statements about observable matters can be. General statements cannot, and no accumulation of instances changes that. What replaces proof is a test that could have failed and did not, reported so that somebody else can run it.

A claim that no observation could contradict cannot be supported by any observation either. Stating what would count against a claim is the single move that turns an assertion into something checkable. It is also what makes it worth citing.