Developer experience (interactivity, debugging), libraries, performance. At least, it is what lispers say: https://gist.github.com/vindarel/c1ef5e043773921e3b11d8f4fe1...

Racket is a single extensive implementation (maybe two versions of itself, the old one and the new one updated with bits from Chez Scheme).

Common Lisp is a language standard with multiple (very different) implementations. SBCL is a native code compiled Lisp with type inference. ECL compiles to C or byte codes, with an interpreter. LispWorks and Allegro CL are commercial Lisps with IDEs and GUIs. CLASP integrates with C++. There are a bunch of other ones.

Common Lisp isn't an academic product.

For what you're missing probably mostly just what is described in the "Escaping the Hamster Wheel of Backwards Incompatibility" section. Racket itself is a more reliable and stable language than many, maybe most, but common lisp is on another level with that shit. It has good and mature libraries for nearly everything, where racket has them only for most things.

Also if you know you ultimately need to deploy a native binary and only a native binary will work, common lisp historically produces alarmingly fast native binaries. I haven't needed to use this in racket much so I can't directly compare; it doesn't seem to prioritize perf but still seems to be much faster than other languages that don't prioritize perf.

I definitely prefer racket though. It's a bit clunky in places too but is overall much better planned and still has all the extensibility benefits.

I would say nothing but also everything. Racket is Schema and Lisp mashed up. Though Common Lisp is very easy to read once you have Racket down. Personally, I prefer Racket for my personal side projects.

Nothing, Racket is specifically designed for education and it will work for many applications. If you hit some wall (lack of libraries/performance issues), you'll be able to pick up common lisp quickly, and be more sympathetic towards its warts and anachronisms.