Tolerance is usually defined as being an allowed variance from a set standard. Just about every profession has some tolerance or ranges of tolerance. For example, buildings are designed to standards which allow certain deflections of floors (Deflection of l/360 is one common tolerance for beam design). The floor can bend by that amount before it is considered a problem.

Projects also have certain tolerances. How firmly fixed is the end date of your project? Is there any leeway one way or the other? If there isn't, there probably should be. The real question is how to determine the appropriate tolerance and to convey it to someone who is less informed about the project than you are. There are people who you give a date to and that date becomes the single acceptable value. They implicitly turn your "plus or minus" into "minus" and allow no plus. What this means is that you have to adjust your target so that even if you reach the limit of your tolerance, you are within the tolerances set by others.

There are a few ways to do this. Theory of Constraints does this with a project buffer. A number of vendors have determined rules of thumb for sizing of these buffers. The problem with them is that they are just rules of thumb. It probably takes a few projects to find out what the real size should be for your project and your team.

Another approach is probabilistic scheduling and simulation. In this approach you build a schedule model and instead of a fixed duration for each task you enter a probability distribution. Then the schedule is calculated based on a large number of random samples. The result is a probability distribution for the completion of your project. If you are successful at this, it gives you a tool which you can use to set a tolerance which you have a high probability of meeting or exceeding. I have a brief article about the basics of this method and a Microsoft Project macro which can introduce you to this approach. You can download it here: Free Download Microsoft Project Monte Carlo Simulation