Re: up front designs always useless

From: Mark Nicholls (nicholls.mark_at_mtvne.com)
Date: 01/18/05 

Date: Tue, 18 Jan 2005 11:04:00 -0000

"Robert C. Martin" <unclebob@objectmentor.com> wrote in message
news:q83ou0lmtls0kqqq22u7p9mv0nk3h1qbb5@4ax.com...
> On Mon, 17 Jan 2005 11:03:18 -0000, "Mark Nicholls"
> <nicholls.mark@mtvne.com> wrote:
>
> >
> >"Robert C. Martin" <unclebob@objectmentor.com> wrote in message
> >news:s03lu05oi58s6ujsgjkvbdsjlk3mr3ge8b@4ax.com...
> >> On Fri, 14 Jan 2005 16:43:52 -0000, "Mark Nicholls"
> >> <nicholls.mark@mtvne.com> wrote:
> >>
> >> >So how much shall I budget for? or are you going to tell me at the end
of
> >> >the project? Shall I tell the share holders that?
> >> >
> >> >This is a central problem for iterative techniques....
> >>
> >> Not at all. Iterative techniques are much more accurate at delivering
> >> this information, simply because they measure reality and continually
> >> refine.
> >
> >the OP is about up front estimates.
> >
> >Thus the a central problem for iterative techniques is on day 1.
>
> This is not a weakness of iterative techniques, it is a weakness
> period. In this position I have two options:
>
> 1. If I have no history of doing iterative projects, then I'll have to
> do a standard task breakdown and estimate. Pert is a good technique
> for doing this. Wide band delphi is another.
>
> 2. If I have a history of doing iterative projects then I have a
> measured velocity. I still have to do a task breakdown, and I still
> have to use a statistical mechanism like PERT, but I can plug in real
> velocity data.
>
> In either case I still wind up with an estimate in the form of a mean
> and variance.
>

Exactly

And in both cases you have made a task breakdown....which implies up front
design decisions have been made....buy? build? 3 tier? thick client?........

And *inherenty* in this is the sense that in order to buy predictability the
project will be overbudgeted......inherently.....in order to buy
predictability from risky parts of the system, I will generally have to
overspec (or what will look like overspecced afterwards) the system in order
to mitigate

You can drive the variance down, but it's always there, and there is still
not a variance box in my CE form....so I must inherently overbudget.....

It's a problem.