depth_point_offsets = pm.Uniform(“depth_point_offsets ”, lower=0, upper=max_depth, shape=num_layers)
depth_points = pm.Determinstic('depth_points', depth_point_offsets.cumsum())
You can do arbitrary transformations of RVs after you’ve made them, so in this case you could just take their cumulative sum. If you want to enforce a maximum, you could pump the depth_points though some bounded function (like inverse_logit).