The software in Markov binomials

 The pit boss software is about the pit boss maintaining its own binomial to balance the two saings and loan binomials.  But the pit boss wants the lowest number of coin tosses, it would want the savings and loans trader to balance skew all by them selves. They can't.  The pit boss has highest priority, Pauli exclusion, spin collisions are a pain. Spin has no entropy and coin tosses are allocated accoding to the binomial that maximizes entropy.

So we have this pit boss dealing with a plt of loans and deposits, in advance. It really finds the Huffman tree for each, and from those two trees, build the smaller tree that will balance net skew. If the pit boss entropy grows greater than zero, it will execute an interest swap, effectively changing the weighting on the coin tosses. This will be a lot like Coinbase exchange matching. 

But in sandbox we have trader bots. So the Huffman encoder is run time adaptive rather than batch, and that  minimizes clocks on the pit boss, it is like a fast transform.  Most of the trader bots will pass, having no information to change S/L. So the Huffman tree is stable to entry and exit on an adaptive basis.

This is doable, and likely done by the information folks already.  This is not rocket science. The automated S and L banker is probably running everywhere folks are doing hidden Markov model. They are doing the same, tracking the proper N from Markov node to Markov node.