She canceled plans the night before. Something came up, she said. You said no problem and made other plans. A week later, in a different conversation, she mentioned that you hadn't tried to see her when she canceled. You thought you were being easygoing. You were failing a test.
The test was real. The grade counted. The rubric was never shared.
The game
She wants to know how you'll actually behave: under pressure, when she's unavailable, when there's an attractive distraction, when something small goes wrong. Direct questions don't give her that. People answer the way they think they should. Observed behavior in a real situation tells her something questions can't.
So she creates situations, or uses existing ones, and watches. She introduces you to her attractive friend and tracks how you act. She picks a minor fight to see how you handle conflict. She cancels plans to see if you'll push back or accept it. Each scenario has a right answer (hers) that she hasn't shared.
You're being evaluated on a rubric you've never seen.
The equilibrium
The test persists as a strategy because covert observation is genuinely better information than self-reporting. People behave differently when they know they're being watched. The methodology has real validity.
The problem is the asymmetry. She's making real decisions about the relationship using information you didn't know you were giving, on criteria you can't adjust for. You can't study for an exam you don't know is scheduled.
Dominated strategies
Trying to intuit what the test was and retroactively pass it. You'll guess wrong, or contort yourself into behavior that doesn't fit, and she'll see that too.
The test is looking for one thing: whether you're different when you think no one's grading you. If you're not, there's nothing for it to find. The ones you failed without knowing you were being tested are already scored.