Regarding the Hubbert linearization on Russia, I don't think this technique is appropriate simply because production has been constrained and the observed profile will be bimodal. Curve fitting is a maximum likelihood approach where there is no prior on the curve parameters and will produce suboptimal results when the residuals are non gaussian.  

If we look at the different reserves estimates you are giving, we can design a rough triangular prior probability density function for the URR based on the different fractiles (F00= 127 + 60, F50= 127 + 113, F100= 127 + 180):

Using a Bayesian filtering technique (or sequential Monte Carlo Markov Chain (MCMC)) we can use the above prior to constrain the solution to fit in a valid subspace of solutions. I can't give all the details of this approach here, but the details can be found on my blog. The predicted production from this algorithm (called SBM-PF) is given on the figure below:

You can also look at the URR values derived from the Bayesian filtering:

The F95 value is 239 Gb (i.e. 95% of chance to have an URR > 239 Gb).

If we compare this Bayesian approach with the standard Hubbert linearization, we can see how the standard Hubbert linearization could potentially underestimate the URR: