The purpose of the size parameter is not well-defined formally, but it is a very convenient knob to easily tune the test suite in various situations, that is definitely worth the negligible cost of having it around unconditionally. Without a size parameter, a fixed distribution means that if we want the generator to cover all possible cases, we have to keep a small probability of generating humongous examples and thus go OOM. We can avoid that by making the size of generated values bounded by the size parameter (or a function thereof), which we can increase easily when more resources become available. Furthermore, if we really want to generate very large examples, the only way with a fixed distribution is to wait longer. Instead, using a size parameter, we can make smaller values less likely to target further regions in the search space more accurately. Some properties can take a while to test on larger examples (either because of the generators or the actual verification process) so we might like to keep the size small during development, and raise it again once we're done. The Arbitrary class assigns a "default" generator for every type. While it is not always a good choice, having a parameter to tweak makes Arbitrary more generally useful. As for your last point, small examples are faster to generate and check, so it seems like a decent strategy to start small by default. Li-yao