[Zigurd]

This is exactly why waterfall works for building a building, and why it doesn't work for building software: You can directly observe progress.

Some of the better aspects of Agile are aimed at providing an analogous level of obviousness: E.g. don't move on until a story is completely implemented. This means you see schedule risk as it happens. This also avoids the way waterfall projects used to die, like the one I saw at Lotus while consulting on Mac porting, back when using conventional project management tools for software was a progressive management practice: A GANTT chart with hundreds of lines corresponding to tasks. Each progress bar partly filled, most of them 70-80%. Yet the project was dead because it was obvious it would not be completed in a relevant time-frame.

It is far too easy to game completion tracking in big, complex projects if you don't decompose them into small sequential projects. BUT, even though each smaller project has schedule risk that is reasonable, when you add up the schedule risk across all the sub-projects, if you are honest, it will be large enough that an estimate is going to have low reliability.