Summary
Mutation survival rate is the fraction of viable mutants in a mutation-testing run that a test suite fails to detect. It is the complement of the more familiar kill rate: a 32.0% survival rate means 68% of mutants were killed. The metric is used to assess test-suite quality, particularly for AI-generated tests, and its interpretation depends heavily on what kind of mutants were generated.
This article synthesizes a personal baseline reading with two external references: SWE-Mutation, an LLM test-suite benchmark, and Erik Perttu’s mutation-testing sufficiency gate. The central takeaway is that survival-rate numbers are only comparable when the mutant generator and test population are similar. A reading is a measurement; a gate is a precondition — and the two are different uses of the same tooling.
Key Points
- Survival rate is not a universal quality score: it depends on the mutant generator. Standard operator-replacement mutants are easier for tests to kill than agentically generated, realism-tuned mutants.
- SWE-Mutation shows that more realistic agentic mutants roughly halve average detection (from 71.04% to 39.
See also: mutation-testing, cargo-mutants, mutation-testing-operator-limitations, greenproof