OK, let's restate a definition meaningful for the purpose of this model (after all, Migeru wanted the thread to be unreadable): oil fields come in two flavors, we have pre-existing undiscovered oil fields, in a finite number  (as you said, they're only produced in geoligical timescales), this number is unknown but the order of magniture can be estimated quite accurately. These are the inputs to produce "known fields" aka reserves, the other flavor, which goes into Migeru's matrix.

Of course, we are not in a pure Leontieff model since the coefficient of oil discovery will vary over time. The oil itself will change also, making refining less efficient, etc...

But this was to be expected: we are modeling a depletion process, which is irreversible just like entropic decay. The pure leontieff model is reversible if you keep all coefficients constants, it is always conceivable to return to a previous state of equilibrium. Which will not be possible with oil: once it's all CO2 in the stratosphere we won't be able to drive so many hummers...

Pierre