There is a gap between finishing a course and being useful on a team, and it is not a knowledge gap. It is a judgement gap. Courses give clean problems with known answers. Real work gives you a half-described requirement, a codebase written by someone who has left, and a deadline.
The skills that close the gap are rarely taught directly. Reading code you did not write. Asking a question precisely enough to get a useful answer. Knowing when a problem is worth another hour and when to raise your hand. Estimating your own work and being wrong in a way you can learn from.
This is the reasoning behind putting learners on live projects with review rather than on simulated ones. The task is scoped small and the review is close, but the work is real and someone is waiting for it. That changes how carefully it gets done.
It also changes what feedback means. A review comment on a practice exercise is an opinion. A review comment on code that goes to a client next week is a decision, and people remember decisions.
The trade-off is supervision cost, and it is genuine. A trainee on live work needs more senior time than a trainee on exercises. We treat that as the price of producing people who can be trusted with the next project, not as overhead to be minimised.