Agreed that the (full range of) inputs are unknown, but without the data being readily available it does make investigation more difficult than it otherwise should be, so I stand by my desire for the data to be made available as easily as possible rather than stuck in proprietary apps that require reverse engineering or laborious data export.Fundamentally this is not a software integration issue, in my view.
The real barrier to the type of predictive engine you seem to be hoping for is that the inputs are largely unknown.
Anyway I fear I've taken this thread off on something of a tangent, my apologies to the OP! 🙂