This is a response to @msa6867 presentation on the node runners call. The presentation touched some sensitive points of this proposal, some of which are not accurate.
You can hear the presentation here:
and see the slides here:
I will try to keep this as short as possible, since the justifications are all in the document provided with this proposal.
TL.DR:
-
Mark talks about boosted rewards. They do exists, but they are wrongly estimated.
-
Multiplying the staked POKT by two means that the rewards are divided by two. Wrong, this is the point that we are proving, it all depends on your QoS. The linear relation does not hold.
-
In slide 13 an example of how the Cherry Picker assigns probabilities is show. This example only analyzes the CP after it has ranked all the nodes in the session.
Finally it was slipped that we had an “agenda”, not sure what he meant by that. We have disclaimed our affiliations and our business is no secret).
What we find misleading is claiming that:
This is not credible, as @msa6867 is the author of PUP-21 which is being replaced if this proposal passes and he probably also owns nodes which are probably staked at the maximum bin (just a guess but it goes in line with he’s arguments).
(Long version now, you can stop reading if you want)
-
Mark talks about boosted rewards due to the reduction of nodes, and that this boost disappears when the PIP-22 was activated, due to the base node multiplier.
This is correct, what we are saying is that the applied correction is wrong. -
In the presentation an example is given, were the duplication of the staked pocket results in a reduction by half of the served relays (regardless the existence of PIP-22 or not). This is the thesis of the linear model that we talk about in section 4.5. The only difference is that instead of a relation of number of nodes and served relays he uses a relation of staked POKT and minted POKT.
We prove that this is wrong.
This conclusion does not take into account how the effects of the QoS on the CP. For this model to be accurate the determination coefficients (D) should be near 1.0, but they are far from it (see table 5 in the main document and table 22 in the appendix).) -
In slide 13 an example of how the CP assigns probabilities is show.
The conclusions are correct, there is no change in probability, this is not our point.
The problem here is that the given example is just a snapshot of how the CP works after it has already measured and ranked the nodes. In real life the CP goes to a transitory state were it measures each nodes. During this process the CP gives each node equal opportunity to serve some relays. None of the models provided by @msa6867 take into account this effect, our simulations did.
Although we cannot fully describe the works of this transitory state of the CP, we can measure its effects. During this transitory state is were low QoS nodes receive relays more often (simple math here). Even when this is a short period of the session its effect is not negligible. More sessions means more transitory states for a low QoS nodes, and hence more gains. This can explain the differences observed in the determination coefficients for low and high QoS nodes (once again, tables 5 and 22). Low QoS nodes served relays is more linearly related to the number of sessions, (but what we want is to be fair with high-QoS nodes, not low QoS nodes.). Finally even if this effect were negligible the difference in the determination coefficients cannot be explained by the model used in the presentation.