E.g., for the CFOP method the second stage is pairing middle edge pieces with bottom corner pieces, and putting them in place. There are 4 of these middle edge pieces, so that stage has 4 steps. Each step may require from zero to ~8 moves.
Speed solvers typically scan the cube for the next step, and then while their fingers execute the moves they scan the cube, looking to setup the next step. Sometimes this allows them to anticipate the next _next_ step.
As an old fart who was a kid when the Rubik's Cube was introduced to America, I'd like these young whippersnappers to solve on the original Cubes, not these Ferrari cubes that never stick or get filled with hand salsa and backpack detritus.
The last layer doesn't really have any decision making though.
The inspection phase is not really about "solving" the cube, but more about checking what is the best route to start solving the cube. In the most commonly used solving method that means that they are checking which color to create the initial cross to and what corners to place in first. If you take out the inspection phase, they will more or less just have to choose that at random, which will only cause more random variation to the final times which is already a huge problem with single solves.
Also comparing current record times by including the inspection time is absolutely useless, as the optimal way has been to use as much of the 15s you have to make sure that you are choosing the right approach and the time used for the inspection has no correlation to the time they have actually needed to inspect the cube.