I like to view it this way: optimal decisions are not computable in practice. We never have enough data, nor enough time, nor is our brain particularly well-suited for this. We have no choice but to approximate, to use heuristics. This applies to both snap decisions and carefully applied formal models. The problem is, even the most rational of us sometimes forget they're just approximating, fudging both data and models at the edges, in order to get some result in finite time. They also forget to mention this.

The difference I see between what I wrote above and what you described as not rational choice of a framework is that, per my personal experience, it is possible to reconcile two conflicting seemingly rational frameworks. It just takes time to extend them past their edges until they meet. That is, extend each past the approximation cut-off point.