Decided that bootstrapping against fixed model parameters would be more difficult to interpret, Fig 1.
Simulated data results with refitting seem strange (Fig. 2), perhaps related to fitting of the tree. Observed LR for data should at least remain symmetric, so something has gone wrong.
Expected pattern should be deducible simply by upper left triangle of graphs, should accept the first model where the data becomes typical.