Reward Curve Fun

in #steem6 years ago

How to interpret the curve in very simple terms.

First things first. Rshares are easy. Posts collect rshares. What you assign to posts is proportional to your SP (for simplicity, assume full voting mana).

Now let’s put the curve on top.

The reward curve assigns a score to each post, based on where they lie on the curve.

The payout distributed proportionally based on the score.

It's as simple as that.

What plotting the derivative of the reward curve gives (as done in the deep dive ) is that it tells you how much additional score you get per rshare if an additional person votes.

Linear is simple, the derivative is flat. No matter where you are in terms of your post score, your score adjusts by the same amount per rshare.

In every other case, you see that the better your post does, the more impact additional rshares have on your score. This is an interesting property that has the following effect for maximizers:

Splitting stake into multiple posts (post farms) is less efficient than consolidating stake into voting a single post, so the preference will be to pile votes into a single post.

Why? It’s very obvious in n^2, but holds true for any reward curve whose derivative is increasing. But as an illustration:

Scenario 1: 10 posts with 10 rshares each : score of 10 * 10^2 = 1000
Scenario 2: 1 post with 100 rshares : score of 100^2 = 10000

And note, your payout is proportional to the score, so scenario 2 is better by a factor of 10!

With n^2/(n+1):

Scenario 1: 10 posts with 10 rshares each: score of 10 * 10^2/(10+1) = 90.9
Scenario 2: 1 post with 100 rshares: score of 100^2 / (100+1) = 99

Here the gap is much less pronounced, in relative terms. Though one thing to note here is that it’s sensitive to the scale. If n here was replaced by n/1000 for example, then we are doing this exercise closer to the ‘quadratic/n^2’ portion of the curve:

Scenario 1: 10 posts with 10 rshares each: score of 10 * (10/1000)^2/(10/1000+1) = 0.00099
Scenario 2: 1 post with 100 rshares: score of (100/1000)^2 / (100/1000+1) = 0.00909

( a factor of 10 again, what do you know )

What this means is, scale is important and can change the system dramatically based on the rshares distribution. If the scale is chosen so that all stakeholders are influencing the earlier, ‘quadratic/n^2’ portions of the curve, then this is really no different than n^2. If the scale is chosen so that all stakeholders can influence on the latter portions of the curve, then this is no different than linear. You can imagine that the scale will be chosen so that we have a gradual transition.

Of course, given the distribution, we know that a handful will be able to push posts rshares to the far right of the curve. (But compare this to the old n^2 system, and just imagine just how much power they had before!)

Anyway, the point is, piling all of one's rshares into few posts is better than massive post farming now. Every post farmer will now need significant stake to escape the quadratic early portion of the curve. Suffice it to say: post farming is now only an option for large stakeholders.

But that's kind of the point of the proposal: as @kevinwong / @trafalgar say (I forgot who said it): bring the rewarded posts out into the light.

And with convergent linear, I'm perfectly happy with it because it doesn't bring them too far out that it's basically at the sun.

There's a downside to this though: it's harder to make money at all at the low end of rshares. Comments may as well not give money at all (except of course if they are exceptional). But I think that's in alignment with how value should be distributed. After these changes it becomes much more important to build your network (which arguably you should be doing now anyway).

So what is the takeaway here? The convergent nonlinear curve looks pretty much like linear except at the low end. The scale needs to be chosen carefully so that it doesn't "curse" too many posts to the quadratic end of the curve. So one need not worry too much about this proposal as long as it is done carefully.

The other good thing about this is that I think the post reward dust threshold (at 2 cents) could even be removed after this curve change and is a less drastic cutoff.

Now… this alone doesn't really do much to address the perceived problems. As mentioned before, self voting without any other change will be for the large stakeholders. Bid bot dynamics is also interesting. Those that use it for promotion will pretty much be operating under the more linear part of the curve. But it will shut out those that are only bidding low total amounts. This again brings "valued posts into the light".

If you continue that quote by the way, I believe it was something like "bring valued posts into the light… …so they can be evaluated further". Meaning, the rest of us can downvote those that are clearly overvalued. Several posts have been made to this effect so I need not explain it further. But requiring varying degrees of free downvotes or increased curation percentage to address various issues.

(Summary: increased curation means increased sniping, which would signal that perhaps the post farmer might opt for doing something else. And more prevalent downvotes will also discourage it further. The same actually applies to bid bots as well, in an indirect manner. )

I hope this can help illuminate a few things regarding this so-called rewards curve.

  • PS: downvotes. Downvotes would also be more powerful when the posts are at the higher end of the curve.
Sort:  

After these changes it becomes much more important to build your network (which arguably you should be doing now anyway).

Exactly. Your blog won’t be more valuable just because of one curve or another.

For the record, I’m happy to change anything that doesn’t undermine STEEM Power as long as we’re honest about the motivation.

The change, as proposed, in theory, maybe, could lead to more demand for STEEM Power. But the same could have been said for n^2.

The same for n2 but wasn't there quite a few large mining accounts already? So wouldn't the buying of steempower need to be quite large and back in 2016 the price was high. So n2 let those miners and their circles/ trails benefit massively with very little risk anyone would threaten their stake.

I wasn't here then but I can't imagine when some accounts mined 5-10% of supply +stinc anyone willing to buy in at that point heavily.

Edit: not my area either so I might not understand :D

Ten accounts take 30% right now.
@statsmonkey
Have been ever since linear was forced on us.

The math worked best when the experiment capped influence at 800mv, imo.
It was threatened that the sky would fall.
Panic set in.

Let them tank the price, they can only sell the ninjamine once.
Steem is here to stay.

My sentiments at the time.

And, dont get me wrong, i fully understand the necessity of how, and presumably why, that played out.
It's just that when steem was open sourced it became a commons.
No longer the personal piggy bank of the creators, but a technological virus, for good or ill, in the hivemind of humanity.

Bring back the n2, cap influence at 500mv (slowly ascending), and see what just one year brings.
Obviously, linear has succeded at supporting its detractor's assertions, and wasted 2+ years on an ego trip, imo.

Sure, n^2/(n+1) seems sensible although im indifferent towards it. I see it doing very little to fix anything.
Killing vote farms? Sure, it could do that. What it wont do is fix that kind of characteristic behavior.

What i cannot stomach is 50/50 which is the big change being talked here. I find that 50/50 will make things many times worse and that there is this delusion present that basically states: "If you increase curation, people will curate more.".

I find that to not be the case at all for many reasons i dont care to repeat again... but heres just one..

Say im a whale....If i really wanted to maximize my profit i can do so under any curve, under any curation split simply by adjusting my content to the minimum work required to avoid downvotes.

Anyways.. I always say i could be wrong, i dont believe i am, but at this point i just hope they will wait with this until i power down my stake.
If this makes things better ill just click the powerup button again, if it doesnt i can run fast.

What i dont understand is that, when faced with such a drastic change more people arent powering down to stay safe. Are you really of such great conviction that such a change on a platform with so many moving parts will turn out positive?
Even if i was in support of this i wouldnt be so sure. :)

As long as it isn't something drastically stupid I'm for it and will influence the new landscape. And you can tell I don't think it's stupid. Compared with the change from n^2 to n, I see this as a return to some nonlinearity without making it crazy at the high end.

Now about 50/50, I see it the following way: it may not make lazy people curate, but if they are just farming they have to wrestle with the nonlinearity AND curation sniping (remember whatever leaked to others in curation in 75/25 doubles in 50/50), and will surely be making less. And this affects bid bot delegation as well. I see that as a win.

Yes your quote is true, but the truth is they actually will be getting less than they otherwise would with no changes. Make even the dumb optimizer gravitate towards behavior that is helpful, or penalize it. As I mention, a post farmer will be encouraged to consolidate their voting to expose their risk to downvotes. But the effectiveness of the system is more contingent upon people actually downvoting. That will be the real test.

Thank you so much for participating in the Partiko Delegation Plan Round 1! We really appreciate your support! As part of the delegation benefits, we just gave you a 3.00% upvote! Together, let’s change the world!

Sometimes curves are always frustrating to study at

True

Hi @eonwarped!

Your post was upvoted by @steem-ua, new Steem dApp, using UserAuthority for algorithmic post curation!
Your UA account score is currently 4.914 which ranks you at #1212 across all Steem accounts.
Your rank has dropped 1 places in the last three days (old rank 1211).

In our last Algorithmic Curation Round, consisting of 205 contributions, your post is ranked at #9.

Evaluation of your UA score:
  • Some people are already following you, keep going!
  • The readers appreciate your great work!
  • Great user engagement! You rock!

Feel free to join our @steem-ua Discord server

Congratulations @eonwarped!
Your post was mentioned in the Steem Hit Parade in the following category:

  • Upvotes - Ranked 10 with 1225 upvotes
 6 years ago Reveal Comment