THE RLM RUBRIC

Lots of things feel close.

Executable environment
Prompt externalized
Code calls the model
Model picks decomposition
State stays symbolic

Plain long-context call
Coding agents + subagents
Hardcoded map-reduce
Recursive Language Model