The mpas-jedi I/O is inherited from MPAS-Models' stream manager. In the typical configuration, mpas-jedi defines four different mutable streams as shown below.

1. background : when reading the background variables for DA or observation simulation.
2. analysis : when writing out (practically overwriting) the analyzed variables from DA. Here the analyzed variables are a subset of background variables.
3. ensemble : when reading the variables for ensemble covariance calculation (of EnVar, NOT EnKF).
4. control : when reading/writing the variables for multivariate B training.

Each stream has a list of predefined variables as a text file (such as stream_list.atmosphere.{STREAM ID}).

In mpas-jedi application, the list of variables for reading is specified in the yaml (JEDI's configuration file). Sometimes when users try adding a new variable for DA, the users add new variable ONLY to the yaml file and NOT to the stream_list.atmosphere.{STREAM ID} file, then having difficulty in figuring it out.

So, I wonder if the list of variables for the mutable stream can be configurable on-the-fly. In mpas-jedi, we may pass the list of variables (which is part of mpas-jedi's State or Increment class) to MPAS-Model's stream manager. Also, mpas-jedi doesn't need to define four different streams. One (e.g., DA auxiliary stream, per @snyderdad) might be enough.