Taleb makes the claim in this paper (and others) that there exists some sort of bound on the variance of a binary forecast such that if a forecaster's binary predictions exceed the bounds on variance there exists a method to arbitrage him. Would someone familiar with the dispute (and perhaps various other papers) be able to explain exactly and technically what he means?
Taleb argues that under uncertainty, election forecasts should be seen as a Binary option. A similar thought is presented by De Finetti's principle that probability should be treated like a two-way "choice" price. Therefore, under high levels of volatility, forecast should not have extreme variation across time (equivalently, the price of the binary option should not change significantly even if polls reveal a large change in the dynamics among candidates). Under high levels of uncertainty, the price of the binary option converges to 0.5. Therefore, the probability of winning the elections in a two-candidate environment should converge to 0.50. A quick look on Silver's forecasts, shows high volatility across time. For instance, Trump's probability of winning the elections ranges roughly from 0.15 to 0.45
For more details on Taleb's no-arbitrage argument, one should check his recent publication and a response to this publication:
Taleb, Nassim Nicholas. "Election predictions as martingales: an arbitrage approach." Quantitative Finance 18.1 (2018): 1-5.