It's difficult to reason about mental models, because we usually equate them to the things they're made to simulate: in everyday speech we say "it's going to rain," not ["my mental model predicts rain"], although the latter is arguably more accurate. Unlike most other species, we have a certain freedom to draw and redraw the boundaries of our black boxes to fine-tune how effectively our mental models simulate reality. Sometimes two framings seem to differ not in where you're drawing the boundaries between black boxes, but in the relationships between those black boxes.