Help with blinding tools in fits

[RENATO_QUAGLIANI]

Dear experts,

I have a very specific problem about blinding mechanisms in RooFit and the proper way to implement it.

My use case and requirements are the following :

I have a model where yields are redefined as a function of 3 RooRealVars [ eff(J), eff(K) , Param]

Yield( species J ) = Yield( species K) * eff( J) / eff( K) * Param

My fit must not show up the eff(J), eff(K), Param fit results.
The fit is also “gauss-constrianing” eff(J) and eff(K) sometimes with a 1D gauss constraint and sometimes with a 2D covariance matrix and of i have eff(J), eff(K) being [0,1] bound always, and this is why I said that the fit must not show up the eff(J),eff(K) results

The final fit results however needs to spit out Param blinded central value, but having unblinded relative uncertainty.

Does anyone can help me to understand if the RooFit BlindTools allow to achieve something like this?

Thanks in advance,
Renato

[RENATO_QUAGLIANI]

Can anyone provide some insight on where to start looking at to implement something like this?
I.e do I need to add functionalities to the currently existing RooAbsRealHidden class or the BlindTools to achieve this?
Apologize @moneta if I ping you directly here, but you might have some advices if this can be achieved already with the existing BlindTools or do I need to implement something new?

[RENATO_QUAGLIANI]

I just want to add that the problem seems to be correctly handled using a custom blinding mechanism for which each efficiency term and R parameter is blinded with a scale transformation being positive defined and independend for each efficiency and R parameter. When handling covariances and gauss constraints we just take the “raw” value and apply the same blinding scale such that

cov(eA,eB) = cov( eAraw,eBraw) * scaleBlind(A) * scaleBind(B) 

The mechanism works only if each term is blinded with a random scale independent for each observable, adn the covariances are also transformed via scales.
As soon as we introduce also an offset, the mechanism breaks.

[RENATO_QUAGLIANI]

It’s however unfortunate that we didn’t manage to make use of any of the BlindTOols available to get what we want.

[moneta]

Hi Renato,

I am not sure if you have solved completely your problem. If you need some extensions or additional functionality for the blinding tools in RooFit, we could also discuss it directly and organise a chat.

Best regards

Lorenzo