[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Measuring Debian progress

On Sun, Jan 07, 2007 at 01:59:35PM -0500, Roberto C. Sanchez wrote:
> On Sun, Jan 07, 2007 at 08:04:31PM +0200, Lars Wirzenius wrote:
> > 
> > * Total number of non-pseudo-packages that bugs.debian.org knows about.
> > * Number of open bugs, in total, and for each severity, and a count of
> > all non-wishlist, non-wontfix bugs.
<snip several useful lines here> ...
> > * Number of bugs opened or closed since the previous report.
> > 
> (Snip mathematical proof of "Bogonic Quality Numbers" :)
> I would like to offer my completely worthless 2 cents here.
> I think that the aggregate statistics are OK.  It will help to identify
> trends in the distribution as a whole.  They are not perfect, but then
> no metric really is.  

Once upon a time, I worked in the employment office, paying people 
unemployment benefit. An office senior manager retired: asked for some 
good advice he said, more or less "Never trust the official figures ..." 
and then recounted that, as a very junior manager in a small office many 
years before, he had been asked to try to fiddle figures for a month - 
to try to get round the system as an official test of how easy it would be to 
defraud/mislead the systems, audits, checks and balances. It was 
surprisingly easy - if you needed to employ young workers under 25, 
you'd concentrate on them to the detriment of everybody else for a week, 
if you needed to show good figures for disabled clients, you'd only 
interview disabled people for a week and postpone interviews with others 
... Needless to say, the example was a good one: just because you _can_ 
prove anything with figures, doesn't mean you _should_ :)

> There are lots of other issues, but I will not go into those now.
> Now, my opinion on the "Bogonic Quality Numbers" is that they should be
> left out entirely.  The only thing worse than metrics is useless or
> incorrect metrics.  Because of the nature of software and the wide range
> of packages in Debian, any attempt to assign a quality metric to a
> particular package is probably not worthwhile.

Anything that will help get an overall picture will help, IMHO :)

Thanks Lars, for some good ideas,

All the best,


Reply to: