You need to know which problem is worth the model. Every rep gives you a real situation, a fixed set of moves, and a rubric you can read the whole way through — then scores the decision you actually made.
Ordered by how much ambiguity you are handed, not by seniority. A director can botch a reframe. Nothing is graded by reading your prose — the answer you build is data, and the rubric is a set of checks over it.
The brief is complete and the data is attached. One defensible diagnosis exists — find it and say why the other readings are wrong.
You stop guessing at causes you can measure.
Constraints conflict and someone loses something. Choose, and defend the choice in a room that disagrees.
You can hold a position under pressure without hiding behind data.
The brief you were handed is wrong. Reframing it is the work — before a single screen gets designed.
You stop shipping well-executed answers to the wrong question.
No brief at all. Here is a place, its numbers and its quarter. Say what is worth building and what you are choosing not to.
This is the job. Everything above is practice for it.
A rep ends in a problem statement. The good ones become entries in the problem library, where developers who have earned the rank take them on and ship a prototype. That is the loop this side of the lab exists to feed.