Sort of like how Clojure or KiCad uses S-expressions to store data. The idea was that like Forth, the parser for such a data structure would be minimal and suitable for even a crappy programmer to implement. I'd thought this would be useful for minimal systems like microcontrollers or locked-down corporate systems with no chance of installing third-party libraries.
float_name 5.75 DUP MUL float
so the transformed data gets loaded in. Could be cool for arrays or more complex data where you want to save the raw data and a record of how it gets transformed.
And I understand you are targeting a minimal implementation, but the concept could also apply to bigger datasets and higher level languages, e.g.
array_name 1 2 3 4 5 "x if x%2 ==0" array