Great. So let’s change this such that the sliced data is in the first argument of the user supplied reducer, the start and end indices go in 2nd and 3rd place.
As a prospect, we should allow for
reduce_sum_propto
as @nhuurre points out. The propto flavour would lead to calls of the unnormalised densities as described (propto=true as template argument given to the user-functor which must be a lpdf or lpmf). That last thing would be a stanc3 only thing to handle, right? This should be done for the 2.24 release given that it is a new feature.
We probably could have caught this earlier, but I really think here the utility of a release candidate shows as now far more eyes are on this. I like it.