An entity that acts on its own schedule also acts while a test is staging the scene

Fact

Simulation tests are usually written as: build the world, place the entities, trigger one action, assert on the result. But any entity that decides for itself when to act — a turret with a cooldown that starts at zero, a spawner, an AI on a timer — starts acting on the first step, which includes the steps the test runs to let the scene settle.

So by the time the action under test is triggered, the entity has already acted once or twice on its own, and the assertion counts all of them.

Why it matters

The failure is asymmetric, and that is what makes it dangerous. A test asserting "this action affects exactly one target" fails, and looks like a bug in the code under test. A sibling test asserting "this action affects at least three" passes — on the extra actions, not on the one it meant to measure. So the pair reads as "one real bug, one healthy test" when in fact neither test is measuring what it claims, and the passing one will keep passing after the behaviour it covers is broken.

It also hides behind a plausible first hypothesis: the obvious suspect is the action's own radius or target selection, which is fine.

How to apply

• Disable autonomy before staging, not after: set the cooldown to something huge before the settling steps, trigger the one action, then set it huge again, because triggering it normally resets the timer.
• Prefer asserting an exact count over a lower bound wherever the mechanic allows one. >= 3 cannot tell three from six, and that is the slack the extra actions hide in.
• When a scene needs entities held still, zero the rate they move by rather than writing their derived position: positions recomputed from a canonical parameter each step will snap back and silently relocate the whole scene.

Related: [[flawless-player-measures-survival]], [[puzzle-tests-must-parse-the-clue]].