The task of judging a library to start with has similarities to the
   task of integrating or composing two libraries.
   Regarding integrating or composing, an interesting example is machine
   learning backends like tensorflow or pytorch. These all have to
   integrate multiple vector math backends, in an organised way, with
   developers likely on a time budget. It's an example of people
   successfully doing this in recent years.
   Regarding judging libraries, the similarity relates to their being
   lengthy lists of api functions, featuresets, or similar, and the need
   to summarise or hold in some way, what is important or possibly
   relevent about these, in order to consider them as a whole with other
   sets of stuff.
   The biggest issue for me with retaining such useful considerations is
   that I lose the parts. Amnesia, disorganization, incineration ...
   Retaining basic small information. Similar to "awesome" lists on gits.
   And using parts of this information to relate with other information,
   make decisions, and do work.