quick answer (believe me I tried), just for clarifications.
First. I agree and share the conclusions that you make, they are the same that we had, we might only differ on how strong their effects are going to be (with GANDALF, we need the rest of the Istar). So, no need to convince me of the benefits of a single chain. Please do not interpret my answers as if I do not share your conclusions.
You are showing only chains that are currently over staked, wich indeed will have increased selection chance when nodes go away from them. The other chains, those under staked right now, is what I’m talking about, those that will gain nodes. Btw, I cannot see that calculation in the provided spreadsheet.
However, lets use that screenshot, if you are a simple staker, would you like to have earning coming at a rate of 34.77% selection prob or a rate of 1.45% selection probability (forgive me gods of math for putting it this way).
For me it is clear that the higher selection rate will result in faster rewards. I have more chance of being selected every hour (session), actually my expectancy is approx 1 every 3. Even when I earn 34 times less in eth than in polygon, I will see rewards every day tricking in. In polygon I will see rewards only approx 1 every 100 sessions, that’s days.
AGAIN, it is not an issue of AVERAGE rewards is an issue of VARIANCE, a variance in time between rewards. Same issue as PIP-22. No, you cannot change this, no it is not unfair, yes it is difficult to explain.
If a chain is left with no nodes (or too little to form a session), PNI will at least spin-up a node ans stake it (probably also hook it up to CC) because it is very likely that they are bound by a contract to do so. So, the issue of starvation has no teeth IMO.
So we are changing this now not to get data but for the thrill of doing it? I don’t see the benefits if we are not going to learn anything from this. If we are not learning anything then it does not matter if we do it before or at V1.
A general note on modeling
Lets try to avoid putting in a higher value to models constructed out of data vs models that are mathematical descriptions. There seems to be an obscurantist trend in the community that calls models “theoretical exercises”, even when they check out with reality. Knowledge building works the other way around, it tries to model things out of simple concepts, compound its effect and test against reality. It is very different from creating model to a fit given observation.
Your model is not “data-driven”, as it relies in a single observation. This is actually a weakness. A data-driven model would be one that explains many observations of an unknown process. You don’t learn nothing from fitting a single observation (or a bunch of them).
As a community we should try to avoid empowering misconceptions like “if it fits the current observation then it is better”. It is much better to have model that we know how and why it works but does not 100% explains everything, than one based “on real dataz” that is only a particular case of completely wrong model.
Don’t get me wrong, I love your threads, they are in fact the highest level of experimentation and research that I see in the forum, thanks for that. Also, this is not a critique to this model in particular, nor a way of saying it is wrong (it is not).
What I need to express is my concern around the way we are constructing knowledge here, because I felt that some of your comments reinforced this misconception.