Yes I am AnonyMint.
I haven't been reading the thread to see if there are any follow ups.
I am still analyzing your design. I think the main issue is because afaics witnesses have nothing-at-stake and afaics it can't be objectively determined which witnesses are creating units referencing MCs which create ambiguity about whether finality was really final. It may also not be objective which witnesses are misbehaving. But I may not yet completely understand the design.
Don't confuse with PoS, witnesses' stakes are
outside the system.
I mean stake in more general terms of what is at risk in the game theory of attacks not specifically its assumed meaning in PoS, but I need to understand your design better before I can posit anything concrete on that. Any way...
I don't understand how you achieve this sentence in Section 6 of the white paper:
We would stop traveling as soon as we had encountered the majority of witnesses.
How can you know what the set of witnesses is in the history when the set is allowed to change in time?
Note I may be wrong about lack of objectivity since I have just noticed your rule that addresses have to reference their prior units in the parent chain else that is objectively misbehavior, so I presume the same rule applies (stringently) to witnesses.
Before I had just skimmed the paper and now I am reading it in detail.
It is the set of witnesses specified in the unit we started traveling from.
The section on 7 on Finality in the Byteball whitepaper is written in a way that I find to be very convoluted. But I was eventually able to understand what you wrote there by conceptualizing it a different way and then relating my conceptualization to your explanation.
My conceptualization is that if witnesses are required to serialize their units and if we can assume we know what all the witnesses can be, then once we have a majority of the witnesses as ancestors of a unit, then the said unit is a stability point because it is impossible for a best parent MC for a future unit to assign a lower ordering number to a double-spend of any unit from which the said stability point unit is a descendant.
But sorry to say the flaw I see in your analysis is that you don't appear to consider the case where a new branch appears (which was formerly hidden by network disruption, propagation delays, or intentionally) which has a sufficiently different set of witnesses from the "current MC" which any particular node analyzed.
Thus afaics, there is no finality. The only way to have finality is to make the set of 12 witness globally static.
P.S. Readers the term 'serial' in the white paper means that it is assumed (how required?) that a witness will only create a new unit (i.e. vertex on the DAG) for which all of the prior units of that said witness have the newly created unit as a descendent. In other words, a witness is assumed to not go creating units which exist on branches of that DAG which don't contain the prior units of that said witness. Without this rule, then the logic of my 2nd paragraph above would not apply.