Stories are the main unit of measurement but we need to keep them small enough to be useful.
Aiming for uniform pieces of work will help it flow through the system more efficiently. The size difference between Smalls and Mediums tend to even out in the wash when measuring velocity so don’t worry about it too much.
TShirt sizing keeps estimation simple. Nobody likes being stuck in an estimation session where we argue the difference between numbers in a fibonacci sequence for 30 minutes.
Stories are a medium by default. Either leave them alone or mark them as Small or Large. You can use relative sizing from story point estimation without the complexity of story points.
Large stories represent significant risk and need to be broken up further (a spike is a great way to do this).
flowchart TD
medium[It's a Medium] --> size{Relative size to other Mediums?}
size --larger--> large[It's a Large]
size --smaller--> small[It's a Small]