Have you read the documentation?

Concave hulls and alpha values

Concave hulls vary in shape. They are calculated by first triangulating the dataset, then measuring the circumradius of the resulting triangles (the radius of a circle that touches all three points of the triangle). Triangles that are larger than the specified Alpha Value are discarded, and the rest are dissolved to form the hull. Measurements are in ground units.

If the Alpha Value is explicitly specified, it is used for the calculation. If it is zero (0) or not specified, the alpha is approximated as the smallest value that produces a single area. If the alpha is too small it may return a null geometry. Larger numbers generally produce larger areas, and a large enough alpha value produces the same result as a convex hull.

See Examples below to see the effect of varying the Alpha Value.

Geometries that are farther away than Alpha Value from the rest of the geometries may be discarded if they do not constitute a valid area on their own.

Concave hull generation works best with points, but line and area geometries will be accepted and converted to points if input.

I thought that was pretty well described in the HullAccumulator documentation. Look at the examples, that shows a hole being created as well. It explains how it achieves the results:

Concave hulls vary in shape. They are calculated by first triangulating the dataset, then measuring the circumradius of the resulting triangles (the radius of a circle that touches all three points of the triangle). Triangles that are larger than the specified Alpha Value are discarded, and the rest are dissolved to form the hull.

Btw: using a DonutHoleExtractor it’s easy to remove the hole.

Edit: Niels posted while I was typing. Sorry, didn’t mean to duplicate answers….

...great minds think alike… is what I think when this happens to me :-)