Last, a model is actually reused (thumb up), but this means, it was found somehow in a larger collection of models. Hence, a component adapts ideas from recommender systems so "the right model" can be found and reused.

Second to last, collections of models where models are actually reused, will expose issues of single models over time. Certainly, these need to be fixed,  discussed, or at least reasoned about. In other words, models evolve or mature over time (king).

First, collections of models are only as beneficial for reuse as they grow continuously and are feed with "good" models. But such models are only found in existing, real world models. Getting the reusable parts out of models preserving a low degree of redundancy it what model harvesting (shovel) should be all about.

Underneath, a persistence layer needs to provide functionality to store, query for, and retrieve models. But this persistence can come of different flavors. This depends on the size of the group working with this persistence (local, or remote, one or few users).

In a nutshell and now from start to end, a model is analyzed for reusable parts, which are harvested, i.e., stored in a model library. Of course, the newly harvested models might not be perfectly suited for reuse. Hence, they evolve in this model library and are offered for reuse after they matured to a certain degree.