Hypothesis

A property-based test describes behavior that should hold for a class of inputs rather than listing only individual examples. Hypothesis draws inputs from composable strategies, runs the test and attempts to simplify a failing example. Shrinking makes a counterexample easier to investigate, while retained examples can help reproduce regressions. The framework does not invent the oracle: the author decides what property should hold. It complements example-based tests and differs from formal proof, because generated exploration is finite and bounded by the supplied strategies and execution settings.

The practitioner selects invariants meaningful to the component, such as round-trip equivalence, conservation of totals or agreement with a simpler implementation. They construct strategies that represent valid structures and deliberately include edge conditions. They avoid filtering away difficult inputs and distinguish unsupported values from implementation defects. On failure, they inspect the reduced example, fix the root cause and retain a focused regression test where helpful. Useful work expands coverage of assumptions while keeping failures understandable and test execution practical.

Weak properties can pass broken code, and unrealistic strategies can spend effort on irrelevant inputs. A round-trip test can miss paired errors when both encoder and decoder make the same mistake. External services or uncontrolled randomness can make results unstable. Check oracle independence, input validity and shrinking behavior. Hypothesis explores the behavior expressed by a test; it cannot establish an unspecified requirement or prove correctness for all possible execution environments.