You got me curious, so I made a little harness comparison to my model test suite:

    Model                              Adapter       Suite           Score   Passed  Tasks
    ---------------------------------  ------------  --------------  ------  ------  -----
    local/ornith-1.0-35b               little_coder  aider_polyglot  36.0%   81/225  225
    local/ornith-1.0-35b               pi_devstack   aider_polyglot  39.6%   89/225  225
    local/ornith-1.0-35b               pi_vanilla    aider_polyglot  32.0%   72/225  225

Little Code does a little better than raw Pi, although maybe not better than my personal Pi setup: https://github.com/lhl/devstack