A user wants to re-link new BGK irfs to replace those in existing observations.

[kosack]

As an example of why this is needed: imagine we have released many DL3 products over a year, and after 6 months, we have created a new background model. We might not want to re-generate all DL3 products and re-release them, but rather just provide the new model for each observation. Then how would the user tell e.g. gammapy to use the new BKG IRFs and not the ones inside the observation DL3 files?

[kosack]

The basic principle could be to use an external grouping table, like the one described in #3, and then modify the values to point to new files.

However, this will be very difficult, since the new model must have all SOIs and event types... not so easy for a user, as that can be a large multiplicity. For something like HESS, where here are no SOIs or even types (so far), it would work similar to how it works in gammapy now. But in the most general case, you would have to generate a background model file per observation.

Then the science tool would have to decide what is the "source of truth" of the IRF grouping table: the external one, or the ones in each observation file (see #1 ).

So this is feasable with the proposed model, but very difficult from a bookkeeping standpoint.