🤖 The AI Prompt That Writes Property-Based Tests for a Pure Function
Example-based unit tests only check the specific inputs you thought to write down, so a pure function with a subtle edge case — an empty collection, a boundary value, an input that’s technically valid but unusual — can pass every test you wrote and still break the moment real data exercises a path nobody considered. Property-based testing flips that: instead of specific examples, you assert a general rule the function must hold for ANY input in a defined domain, and a property-based library generates hundreds of cases to try to break that rule — but writing good properties by hand takes a different kind of thinking, and a blank page is where most people give up on the technique.
🔎 The Problem
Here is a pure function. Write property-based tests for it using <FRAMEWORK, e.g. fast-check / Hypothesis / FsCheck>. 1. Identify the function's invariants - properties that must hold for EVERY valid input, not just the examples I give you (idempotence, a round-trip with its inverse, a relationship between input size and output size, commutativity). 2. For each invariant, write one property test with a clearly named generator describing the input domain, biased toward edge values (empty, zero, negative, max-length), not just uniformly random ones. 3. Do not write example-based assertions (expect(f(2)).toBe(4)) - every assertion must be a property that holds across the generated range. 4. List, in a short comment above each property, what specific bug class that property is designed to catch. 5. If the function has a precondition (sorted input, non-negative, etc.), encode it in the generator itself, not as a runtime check inside the test. Function: <PASTE THE FUNCTION>
✅ Why This Works
- Requiring a named generator per property forces you to think about the function’s actual input domain before writing any assertion, which is usually where the real insight in property-based testing happens.
- Banning example-based assertions keeps the model from quietly falling back to the easier, weaker style it defaults to when a property is hard to express.
- Asking for the bug class each property targets turns the test suite into documentation of what could go wrong, useful long after the session that generated it is gone.
🚨 Why This Is Dangerous
- A property-based test can pass for years and then fail once on a generated input nobody saved, because most frameworks don’t persist the failing seed by default — always capture and commit the failing case as a permanent example-based regression test the moment one turns up, don’t just rerun and hope.
An example test tells you the function works on the inputs you thought of; a property test tells you it works on the ones you didn’t.
