Differential testing uses one implementation, model or specification as a comparison point for another. When both receive the same inputs, a disagreement identifies a case that deserves investigation even when the expected answer was not written in advance.
The method is useful when translating verified models into production languages or comparing optimized implementations. Agreement is evidence, not a proof, because both systems can share the same mistake or fail on cases that were never generated.
Acronyms and aliases
differential software testing variant
Related terms
Frequently asked questions
What does differential testing compare?
It compares outputs, errors or state changes from multiple implementations that receive the same generated or curated inputs.
Is differential testing a form of formal verification?
No. It is a testing technique based on sampled executions, although it can compare production code against a formally modeled reference implementation.
