[Rivet] [Rivet-svn] r2908 - trunk/src/Analyses

[Frank Siegert]

Hi Andy,

On 02/02/11 13:55, Andy Buckley wrote:
> It had been validated and then I rewrote it to clean up: we used that
> analysis as an implementation exercise exercise at YETI. I knew there
> was a remaining problem with the histograms -- thanks for finding it --
> but as none of the analysis logic had changed I hadn't marked it as
> UNVALIDATED.

 From a quick look with a few Sherpa events this analysis doesn't look 
correct though, and the differences are so big (factors of 100 in the 
dijet mass spectra, 2 in some of the jet pTs, completely off for the 
dijet chi) that I tend to think it's the analysis which is wrong.

> While not automatically a fan of formal procedures, I think we do need
> to introduce a more formal process for analysis validation. Running lots
> of MC and regression testing before each release is one thing, but I
> think that we should require supplied analyses to be code-reviewed by
> experienced developers before they can be marked as VALIDATED in the
> first place. That's more important as we start to get more analyses
> direct from the experiments -- several have been spaghetti code that
> achieve their objectives in extremely silly ways. We should aim to keep
> the standard and style of code consistently high quality.

I agree, but am not sure how to avoid us having to write all analyses 
ourselves in the end.

By the way: With regards to validation of Atlas analyses I've found it 
very useful to run the Rivet_i interface in Athena over the same MC 
samples which have been used in the publications. Of course this doesn't 
solve the issue of code review though.

Cheers,
Frank