 Known-answer test

A known-answer test checks a cryptographic implementation by supplying fixed, completely specified inputs and comparing its output with an independently established expected value. The input and expected output together form a cryptographic test vector; the test is the act of executing the implementation and checking that agreement. Published HMAC vectors explicitly support conformance testing.^[ref]

For example, take the four key bytes 4a 65 66 65, representing Jefe, and the 28 ASCII message bytes representing what do ya want for nothing?. The published HMAC with SHA-256 result is

5bdcc146bf60754e6a042426089575c75a003f089d2739839dec58b964ec3843.

A known-answer test decodes that 64-digit hexadecimal value to 32 expected bytes, computes the keyed result, and compares all 32 bytes. Neither the message's terminating string-storage zero nor a newline is part of these input bytes.^[ref]

Several different errors can cause failure: decoding hexadecimal as text, including an extra byte, using the wrong key, selecting another cryptographic hash function, or serializing the result incorrectly. A failure establishes disagreement with the vector; it does not identify the responsible step by itself. A useful test harness reports the selected function, input lengths and first differing output position.

Passing also has a precise limit. A defective implementation that returns this expected value for this one input and zero for every other input passes this test. Thus a finite set of known-answer tests is evidence about those cases, not a proof of correctness for all inputs or a proof of cryptographic security. Boundary cases complement ordinary short examples: the primary vector set includes output truncation and keys longer than a hash's input block.^[ref]

An independently fixed expected value is essential. Computing the “expected” output with the same implementation under test can reproduce the same defect on both sides. Known-answer tests are especially useful for reproducible checks of exact byte conventions across different implementations.