My ideal reform would be the Land Value Tax[0], which admittedly has run into problems in execution with poor implementations.
It depends upon assessments on land value that reflect current market value of that land. In one notable case of the LVT in practice (Pittsburgh), assessors made a real hash of it, with assessments well out of proportion to market demand.
I would like to see a proof of concept to fix this, mainly an automated algorithmic assessment map of land values, based off of public sales records. It's a fun project to attempt; I should start up a repo some time.
[0] https://en.wikipedia.org/wiki/Land_value_tax