A graph can never be better than the data it's based on. To be able to trust the burn up/down graph you need to be sure that what is said to be done really is 100 percent done and not just almost done. If work is remaining but hidden your prediction will never be correct. The tool for this is to specify the definition of done so that when someone says they are done it's clear for everyone what has been done and what is left to do. The other tool is to have short feedback loops with continuous verification.

It's very hard to predict how much work there will be to correct problems found during a performance test or an acceptance test before you have done the test. The result could be that everything is clear or that you need to redo the whole work. The following graph is a typical symptom of hidden work that is kept to the end of the project:

The typical burn down graph for teams that keep testing, documenting, and deploying to the end of the project

With the continuous...