Frozen test fixtures can create significant problems in large software projects. As test suites grow, the more developers rely on shared fixtures, the higher the chance that changes to these fixtures will lead to false test failures. Each test often makes implicit assumptions about the fixtures, and when those assumptions are broken, tests can fail even when the code is functioning correctly. This issue becomes particularly severe when dealing with thousands of tests, where modifying fixtures can lead to many unrelated tests breaking, resulting in a hesitancy to change them at all. The author outlines two common but ineffective strategies used to manage this problem. One approach involves creating new fixtures whenever existing ones can't be reused, which leads to an ever-expanding collection of fixtures that complicates understanding and maintenance. The second strategy is to modify fixture records directly within test code. While this may seem straightforward, it often leads to an ad-hoc solution that complicates the testing process. Instead, the author advocates for a disciplined approach where tests focus solely on the specific properties they are meant to validate. This method requires careful consideration and practice but can alleviate the freezing effect on fixtures. Concrete examples illustrate how to write better tests. For instance, instead of asserting that a specific set of projects is returned, it’s more effective to check that the returned collection includes certain active projects without being affected by the addition of new ones. Similarly, when testing collection order, rather than checking for a specific sequence, tests should confirm that the collection is sorted correctly. These strategies help maintain the integrity of the test suite while minimizing dependencies on fixtures. The article concludes by addressing the ongoing debate between using fixtures and factories in testing. The author takes a pragmatic stance, suggesting that both have their merits depending on the project's needs. Instead of choosing one over the other, combining both can often yield the best results.