Fossil data, although useful to evolution, presents a conceptual problem for evolution. Fossil data is discontinuous data and therefore will lead to a discontinuous theory, by default. In other words, if the process of evolution was continuous, for the sake of argument, but only partial and discontinuous data existed, one may not be able to prove continuous, but rather the data would support a discontinuous theory.
As an analogy, which you can do as an experiment, make a complex shape with popcorn in a field somewhere. Allow some time to pass so the birds, bugs, and little animals eat away at the shape until all there is left is 1% of the original shape. One may not be able to infer the original shape from what is left, since it is only a random and broken image of itself. Instead the remaining popcorn data may be better defined by a different shape that is false to the original image.
Another way to test this is to use modern real time data of life that is fully stocked with all the options so all nuance is side by side. With humans there are large, small, tall, short, strong, weak, different colors, birth marks, hair colors and textures, etc. Fossils data would only have a few of these dozens of options, and will assume this limited set was selected. But the real time data that has all these options selected since all are here. Since there are too many selections, to call it selection, maybe another theory would be needed. But with fossil data narrowing down the options in an arbitrary or random way, the idea of a selection process appear to better describe the limited data.