TL/DR version:

1. We made executable simulation of multibranch forging

2. We think multibranch forging properties are like a lot Transparent Forging properties

3. Multibranch forging results are looking much better than single-branch: nice Gaussian distribution around unit. In practice it means e.g. 1 min average gap with max. 4 mins gap max in very rare cases.

4. Computational efforts still an open question, with big prediction depth we are getting proof-of-work instead of proof-of-stake again.

5. Some things in our executable model are proved formally with the help of Coq theorem prover. E.g. it's proven formally our that our forging trees are balanced right

P.S. With multibranch forging it's easy to define & show nothing-at-stake problem. Next paper will be about that

P.P.S. Also our understanding of consensus in proof-of-stake networks is much deeper now, after playing with simulations(we have another one, Nxt-like, which will be presented soon as well)