Selfish Mining Fallacy - Debate IV

csw

@elliotolds "csw apparently doesn't realize that the fraction of miners on the selfish block is meant to represent how important the "head start" is of the honest miners. Think about it: if the head start is super important, it means that most other miners will be mining on the honest block because they've heard of it first. So gamma (the proportion of miners mining on the selfish block) will be low, like 0.05, or even 0). The gamma parameter is a distillation of how important this "head start" is."

No, I do realize.
Wrong model

"It seems like most of the non-technical objections to the selfish mining paper are "but.. the honest miner has a head start! Doesn't that mean that the honest block will usually be accepted first?" This is the most basic intuition one could have about why the selfish mining paper doesn't work, and I assure you, everyone knows about this objection. The reason the algorithm in the paper is cool is that it shows that despite the fact that the honest miners have the head start advantage, selfish mining still works because other aspects of the algorithm more than compensate for that. Selfish mining still works (if the selfish miner has over 1/3 of the hash power) if the honest miners win every single race when the selfish miners try to release a block slightly after the honest miners."

No, you have no idea about the system.

But this has been valuable.

What I see, is nobody here is running a miner of any scale. I would say wallets at best.
You are confused and think wallets do more... A misunderstanding that I have the data do shoot down now that I know the misconception.

@elliotolds
Yes, Conditional probability - a shame people do not see the model flaw. Oh well. You will.

"the blue text is just a restatement that gamma is low (aka, the selfish miners being a step behind is really important).. but as I've explained, this doesn't change anything in the paper"

csw [11:10 PM]
@xhiggy Now you are starting to think!

csw [11:12 PM]
If you knew how the network works and the topography, then you would say this.

The toss up is 100% of Miners act ONE way
Gamma Miners react another.

" The problem is that the number of honest miners they can attract is essentially zero "
They have a chance to gain up to 0.1% of the honest nodes. But this is the real world. Not their model

Grouping is standard in random systems...
But... SM stands out like a sore thumb. This is right, nobody ACTUALLY modeled this using systems...
You all just plug things into the equations given and not run a real simulation.
If anyone had actually run this on real systems (testnet or their own VMs) as a real experiment, they would have seen the way this all works in reality.

Revenue.

@peter_r Not Profits

Revenue

peter_r
Which is why the selfish miner needs to keep his strategy going for a long time after the difficult resets to bring his profits back up.
Wait till they get ahead...

cypherblock
the paper could be clearer on its claims. ‘not incentive compatible’ , ‘revenue larger than their fair share’. Is it expected that sm strategy leads to greater revenue (ignore profits for now) than honest mining strategy?

csw [11:21]
Let us see....

SM 1/3
HM 2/3

SM gets a block every 30 min on average
HM every 15 mins

Wait to catch up....

I guess that I should say that the model has made a simplification that is common when the results are truly independent. That is they model a set average and not the distribution.
Where the results are independent, this is ok with respect to a Poisson process.
Where there is any dependency it is not.

zbingledack [11:32 PM]
I don't know if this has any relevance, but wikipedia notes, "For any trial of [coin] tosses, it is twice as likely that the triplet TTH will occur before THT than after it. It is three times as likely that THH will precede HHT, than that THH will follow HHT." Seems pretty counterintuitive.

csw [11:32 PM]
The probability of 7 black cards in a row in a shuffled deck is about 1/3rd
Most people do not get probability
When you have conditional probability, there is an intersection of events that needs to be subtracted from the calculation...
That is not down in the equations.

Here is the question for the day....
Being that Selfish miners always reject blocks from others when they have their own (and release competing ones INSTANTLY) what is the pattern in orphan blocks in an attempted SM strategy?

Before it was ASSUMED as with too many things in that paper... Did anyone test this?
And ...
What is the distribution between two competing Poisson processes under conditionality....?

Did we all assume that we can just use averages... Well, yes, the model did.
In a world of ONLY fees, it is even more fun.
Does nobody do science anymore?
You know, make a hypothesis and then TEST it
Not test the model, test the system against the model

zbingledack [11:45 PM]
Still thinking. The block race as a Poisson process is very strange to start with. I don't think I have fully wrapped my head around even the basic design, let alone additional issues. We don't even ever really know a miner's hashpower except for statistics over a given time interval.
I am one who likes to solidify the basics to ridiculous degree before moving to advanced topics. So I will be slow to get these higher-level arguments.

csw [11:50 PM]
Yes.
Assuming the uniform distribution as is done in independent events...

• What is probability of one poisson process occurring after another given that the first occurred?
• What is the probability of the HM discovering 30 seconds after the SM?
• What is the probability that the SM will have a block mined given that they have seen the HM release a block?
  These ARE missing parts of the model.

@zbingledack
Let us start now with the FIRST block event.
A SM finds a block. They Hide it.
If they release, they have a reward of 1 (close to 100% chance)
If they hide it, they have 0 reward but a later possibility.

Chances are (and I will go over the distributions after) that the HM will find a block before the SM gets a second (I will cover this later).

So, they have 100% getting paid as an HM or a CHANCE to risk it. What is the Probability that the SM will lose when the HM discovers a competing block?
P(SS | Not.H) vs P(S|H) = reward.
P(H|S) = they lose....

@zbingledack
Without needing to go into the actual distributions....

Starting to see why this is a conditional probability calculation...?
I am not running the SM state system, I am running Bitcoin. Who cares what SMCoin does...
If you miss the conditionals in the probability calculations you have a value greater than the base and an error.

mwilcox [12:57 AM]
hmm - surely not 100% as HM if SMs have chance to overtake?

csw [1:36 AM]
SM is slower - always under 50% remember
It is not the chance of finding a block, it is for the HM a chance to find a block after the SM has found one...
And did anyone model 2 or more SMs?