But we do have a good grasp of how much is left to be discovered.  The discovery data is well ahead of production.  If we tie the discovery data into production, this should be factorable in as a variance percentage.  

Actually to a degree the Hubbert model does try and account for oil that is yet to be discovered.

In Deffeyes 2005 book "Beyond Oil", he applies Hubbert linearization not only to the oil we have already used, but also to two others.

The second plot is the amount of oil we have used plus the known reserves (removing bogus reserves from OPEC nations).

The third plot is 'hits'.  Essentially a measure of oilfield discoveries, and this is an attempt to measure the amount of oil in oilfields yet to be discovered.

You can then fit straight lines to all 3 of these, and he has such graphs in one of his books.  In the summary, he has the parameters for the lines that he fit to the things.  All 3 have a Qt of 2.013 trillion barrels.

The first curve (oil that we have consumed already) has a=5.9%, peak 2005, and percentage used is 49%.

The second curve (oil that we have consumed plus reserves) has a=7.2%, peak 1978, and percentage of total of 82%.

The third curve (hits - oilfields that we know about) has a=8.1%, peak in 1964, and percentage of total of 94%.

The third curve implies that only 6% of the total world oil is in oilfields that we don't know about yet.

I have to admit that I am still a little fuzzy on the concept of 'hits', and what it means.