Alexander Held

I had not noticed `--dump-datacard` before, but this looks super useful. Thanks a lot!

The question of what to include comes down to who the target audience for this package is (apologies if that has been discussed previously elsewhere!). Are we aiming to provide...

This looks like it could be a convergence issue similar to #1278. In that case a temporary solution may be lowering the POI bound from `0` to something like `-5`...

This sounds like you may have `iminuit

It sets that parameter `mu_BG` to constant by default, so it is no longer a free parameter unless explicitly requested (`model.config.suggested_fixed()` will have it as fixed by default).

For debugging purposes, here are likelihood scans: ```python import numpy as np import pyhf import matplotlib as mpl import matplotlib.pyplot as plt spec = { "channels": [ { "name": "ch",...

The maximum for the Poisson `staterror` case in the example above is at 1000/150 = 6.67 (data / staterror), which looks a bit suspicious. Is there something wrong with the...

Looking what actually goes into the final Poisson call in the backend: for a `shapesys`, the rate is the parameter times auxdata (which itself is (1000/150)^2). For the `staterror` with...

I think this idea is great in theory, and in practice would work best with a gigantic user base where the counts are not mostly going to be driven by...

As described in the [HistFactory documentation](https://cds.cern.ch/record/1456844) in section 3.2, there are two conventions user may follow, resulting in `auxdata=1` or `auxdata` equal to the luminosity. This may be sufficient motivation...