The best roadmap would include a number of increasingly complex result, that we can achieve as fast as possible, each time showing the effectiveness of the method.
The main idea is to solve issues one by one, instead of tackling immediately the full complexity of the problem.
Central value + uncertainty to generate a multi-Gaussian likelihood.
- data loading ->
lhapdf - no theory (it's just a delta)
- simple comparison
Closure Tests like: underlying PDF, known theory.
- data loading ->
lhapdf+ theory - pseudo-theory: whatever theory we use to generate will enter in the likelihood
- again simple comparison: we still have an underlying PDF
- it will require validphys: central values + experimental covariance matrix
- doable the standalone loading of central values, but covariances might be complicated to construct manually