Mutation testing answers one question: change this line, does an assertion notice. That is a fact about the coupling between a test and the code it names. Whether the test drives that code the way production drives it is a separate question, and a mutation cannot reach it. When the fixture is wrong, the mutated and unmutated runs fail for the same wrong reason, the evidence looks complete, and the fix never executes on the path it was written for [13].
The receipt here is two line numbers. In FishNet's `ManagedObjects.cs`, `Deinitialize(asServer)` is called at 418 and `UnityEngine.Object.Destroy` on the GameObject at 437 [9], nineteen lines later in the same file [1]. `Deinitialize` flips the flag through `SetInitializedStatus(false, asServer)` at `NetworkObject.cs:1147` [10], and `DecidesHere()` reads that flag [8]. Unity runs `OnDestroy` after the object has stopped believing it has authority. "My fixture was more cooperative than reality, in exactly the spot the feature lived," the developer writes in the Curbside devlog [20].
Driving the real path turned out to be the cheap part. `StartServer()` then `StopConnection(true)` then `StartServer()` in a single process is 180 lines, runs in about 40 seconds, and uses a launcher another test already used [15]. It had been written off as flaky process-management engineering [16], which is the usual reason a lifecycle test does not exist. Once the real lifecycle ran, the failure was immediate [18]. Ahead of that sat a fortnight of debugging [19] and two consecutive fixes that were green, mutation-proved and inert [14].
For this to be your bug, you need a framework that mutates its own state during teardown, a guard that reads that state from a destruction callback, and a fixture that answers the guard from a constant. Unity with FishNet supplies the first [9][10]. Any hand-rolled arbiter stub supplies the third [12].
The loop is worth keeping on this evidence. The same mutation standard caught a mechanic that shipped with its entry point having no caller under a fully green suite, with every rule of the feature tested and nothing in the game able to reach it [4]. Reading the tests would never have surfaced that. What the standard cannot price is fixture fidelity, so it belongs on the assertion and something else has to own the stub. And the evidence base is one developer, one game still in grey boxes, one networking library [1]. The mechanism generalises, because the teardown order lives in the library rather than the game, but the frequency does not, since there is no denominator.
