Ventolin, I don't mean to come across as too abrasive, but what the hell. Let me spell out why your numerical-predictive posts come under fire: applying statistical models to actual data to make predictions is fine. Your approach, however, always involves making some (fairly kind) assumptions about what our athlete in question could "really" run for their weak distances. I'm not contesting the fact that Tadese could likely run faster than 12:59; in fact, I'm nearly certain he could.
The point is, though, that you're making assumptions about times and then running these assumptions through a model (based on assumptions) to come up with a guess at how fast he could run. That is ridiculous. If anything, your trials that produce 2:01 - 1:58 marathons suggest that either the model is weak or your assumptions are too generous.