Dear STEEM community,
Following our announcement for DTube v0.9, I have received countless questions about the new blockchain part, avalon. First I want to make it clear, that it would have been utterly impossible to build this on STEEM, even with the centralized SCOT/Tribes that weren't available when I started working on this. This will become much clearer as you read through the whole wall of text and understand the novelties.
SteemPeak says this is a 25 minutes read, but if you are truly interested in the concept of a social blockchain, and you believe in its power, I think it will be worth the time!
MOVING FORWARD
I'm a long time member of STEEM, with tens of thousands of staked STEEM for 2 years+. I understand the instinctive fear from the other members of the community when they see a new crypto project coming out. We've had two recent examples recently with the VOICE and LIBRA annoucements, being either hated or ignored. When you are invested morally, and financially, when you see competitors popping up, it's normal to be afraid.
But we should remember competition is healthy, and learn from what these projects are doing and how it will influence us. Instead, by reacting the way STEEM reacts, we are putting our heads in the sand and failing to adapt. I currently see STEEM like the "North Korea of blockchains", trying to do everything better than other blockchains, while being #80 on coinmarketcap and slowly but surely losing positions over the months.
When DLive left and revealed their own blockchain, it really got me thinking about why they did it. The way they did it was really scummy and flawed, but I concluded that in the end it was a good choice for them to try to develop their activity, while others waited for SMTs. Sadly, when I tried their new product, I was disappointed, they had botched it. It's purely a donation system, no proof of brain... And the ultra-majority of the existing supply is controlled by them, alongside many other 'anti-decentralization' features. It's like they had learnt nothing from their STEEM experience at all...
STEEM was still the only blockchain able to distribute crypto-currency via social interactions (and no, 'donations' are not social interactions, they are monetary transfers; bitcoin can do it too). It is the killer feature we need. Years of negligence or greed from the witnesses/developers about the economic balance of STEEM is what broke this killer feature. Even when proposing economical changes (which are actually getting through finally in HF21), the discussions have always been centered around modifying the existing model (changing the curve, changing the split, etc), instead of developing a new one.
You never change things by fighting the existing reality.To change something, build a new model that makes the existing model obsolete.
What if I built a new model for proof of brain distribution from the ground up? I first tried playing with STEEM clones, I played with EOS contracts too. Both systems couldn't do the concepts I wanted to integrate for DTube, unless I did a major refactor of tens of thousands of lines of code I had never worked with before. Making a new blockchain felt like a lighter task, and more fun too.
Before even starting, I had a good idea of the concepts I'd love to implement. Most of these bullet points stemmed from observations of what happened here on STEEM in the past, and what I considered weaknesses for d.tube's growth.
NO POWER-UP
The first concept I wanted to implement deep down the core of how a DPOS chain works, is that I didn't want the token to be staked, at all (i.e. no 'powering up'). The cons of staking for a decentralized social platform are obvious:
- complexity for the users with the double token system.
- difficulty to onboard people as they need to freeze their money, akin to a pyramid scheme.
The only good thing about staking is how it can fill your bandwidth and your voting power when you power-up, so you don't need to wait for it to grow to start transacting. In a fully-liquid system, your account ressources start at 0% and new users will need to wait for it to grow before they can start transacting. I don't think that's a big issue.
That meant that witness elections had to be run out of the liquid stake. Could it be done? Was it safe for the network? Can we update the cumulative votes for witnesses without rounding issues? Even when the money flows between accounts freely?
Well I now believe it is entirely possible and safe, under certain conditions. The incentive for top witnesses to keep on running the chain is still present even if the stake is liquid. With a bit of discrete mathematics, it's easy to have a perfectly deterministic algorithm to run a decentralized election based off liquid stake, it's just going to be more dynamic as the funds and the witness votes can move around much faster.
NO EARLY USER ADVANTAGE
STEEM has had multiple events that influenced the distribution in a bad way. The most obvious one is the inflation settings. One day it was hella-inflationary, then suddently hard fork 16 it wasn't anymore. Another major one, is the non-linear rewards that ran for a long time, which created a huge early-user advantage that we can still feel today.
I liked linear rewards, it's what gives minnows their best chance while staying sybil-resistant. I just needed Avalon's inflation to be smart. Not hyper-inflationary like <HF16, but not leading to a slow death like the current real-world settings of STEEM's inflation.
The key metric to consider for this issue, is the number of tokens distributed per user per day. If this metric goes down, then the incentive for staying on the network and playing the game, goes down everyday. You feel like you're making less and less from your efforts. If this metric goes up, the number of printed tokens goes up and the token is hyper-inflationary and holding it feels really bad if you aren't actively earning from the inflation by playing the game.
Avalon ensures that the number of printed tokens is proportional to the number of users with active stake. If more users come in, avalon prints more tokens, if users cash-out and stop transacting, the inflation goes down. This ensures that earning 1 DTC will be about as hard today, tomorrow, next month or next year, no matter how many people have registered or left d.tube, and no matter what happens on the markets.
NO LIMIT TO MY VOTING POWER
Another big issue that most steemians don't really know about, but that is really detrimental to STEEM, is how the voting power mana bar works. I guess having to manage a 2M SP delegation for @dtube really convinced me of this one.
When your mana bar is full at 100%, you lose out the potential power generation, and rewards coming from it. And it only takes 5 days to go from 0% to 100%. A lot of people have very valid reasons to be offline for 5 days+, they shouldn't be punished so hard. This is why all most big stake holders make sure to always spend some of their voting power on a daily basis. And this is why minnows or smaller holders miss out on tons of curation rewards, unless they delegate to a bidbot or join some curation guild... meh. I guess a lot of people would rather just cash-out and don't mind the trouble of having to optimize their stake.
So why is it even a mana bar? Why can't it grow forever? Well, everything in a computer has to have a limit, but why is this limit proportional to my stake? While I totally understand the purpose of making the bandwidth limited and forcing big stake holders to waste it, I think it's totally unneeded and inadapted for the voting power. As long as the growth of the VP is proportional to the stake, the system stays sybil-resistant, and there could technically be no limit at all if it wasn't for the fact that this is ran in a computer where numbers have a limited number of bits.
On Avalon, I made it so that your voting power grows virtually indefinitely, or at least I don't think anyone will ever reach the current limit of Number.MAX_SAFE_INTEGER: 9007199254740991 or about 9 Peta VP. If you go inactive for 6 months on an account with some DTCs, when you come back you will have 6 months worth of power generation to spend, turning you into a whale, at least for a few votes.
Another awkward limit on STEEM is how a 100% vote spends only 2% of your power. Not only STEEM forces you to be active on a daily basis, you also need to do a minimum of 10 votes / day to optimize your earnings. On Avalon, you can use 100% of your stored voting power in a single mega-vote if you wish, it's up to you.
A NEW PROOF-OF-BRAIN
No Author rewards
People should vote with the intent of getting a reward from it. If 75% of the value forcibly goes to the author, it's hard to expect a good return from curation. Steem is currently basically a complex donation platform. No one wants to donate when they vote, no matter what they will say, and no matter how much vote-trading, self-voting or bid-botting happens.
So in order to keep a system where money is printed when votes happen, if we cannot use the username of the author to distribute rewards, the only possibility left is to use the list of previous voters aka "Curation rewards". The 25% interesting part of STEEM, that has totally be shadowed by the author rewards for too long.
Downvote rewards
STEEM has always suffered from the issue that the downvote button is unused, or when it's used, it's mostly for evil. This comes from the fact that in STEEM's model, downvotes are not eligible for any rewards. Even if they were, your downvote would be lowering the final payout of the content, and your own curation rewards...
I wanted Avalon's downvotes to be completely symmetric to the upvotes. That means if we revert all the votes (upvotes become downvotes and vice versa), the content should still distribute the same amount of tokens to the same people, at the same time.
No payment windows
Steem has a system of payments windows. When you publish a content, it opens a payment window where people can freely upvote or downvote to influence the payout happening 7 days later. This is convenient when you want a system where downvotes lower rewards. Waiting 7 days to collect rewards is also another friction point for new users, some of them might never come back 7 days later to convince themselves that 'it works'. On avalon, when you are part of the winners of curation after a vote, you earn it instantly in your account, 100% liquid and transferable.
Unlimited monetization in time
Indeed, the 7 days monetization limit has been our biggest issue for our video platform since day 8. This incentivized our users to create more frequent, but lesser quality content, as they know that they aren't going to earn anything from the 'long-haul'. Monetization had to be unlimited on DTube, so that even a 2 years old video could be dug up and generate rewards in the far future.
Infinite monetization is possible, but as removing tokens from a balance is impossible, the downvotes cannot remove money from the payout like they do on STEEM. Instead, downvotes print money in the same way upvotes do, downvotes still lower the popularity in the hot and trending and should only rewards other people who downvoted the same content earlier.
New curation rewards algorithm
STEEM's curation algorithm isn't stupid, but I believe it lacks some elegance. The 15 minutes 'band-aid' necessary to prevent curation bots (bots who auto vote as fast as possible on contents of popular authors) that they added proves it. The way is distributes the reward also feels very flat and boring. The rewards for my votes are very predictable, especially if I'm the biggest voter / stake holder for the content. My own vote is paying for my own curation rewards, how stupid is that? If no one elses votes after my big vote despite a popularity boost, it probably means I deserve 0 rewards, no?
I had to try different attempts to find an algorithm yielding interesting results, with infinite monetization, and without obvious ways to exploit it. The final distribution algorithm is more complex than STEEM's curation but it's still pretty simple. When a vote is cast, we calculate the 'popularity' at the time of the vote. The first vote is given a popularity of 0, the next votes are defined by (total_vp_upvotes - total_vp_downvotes) / time_since_1st_vote. Then we look into the list of previous votes, and we remove all votes in the opposite direction (up/down). The we remove all the votes with a higher popularity if its an upvote, or the ones with a lower popularity if its a downvote. The remaining votes in the list are the 'winners'. Finally, akin to STEEM, the amount of tokens generated by the vote will be split between winners proportionally to the voting power spent by each (linear rewards - no advantages for whales) and distributed instantly. Instead of purely using the order of the votes, Avalon distribution is based on when the votes are cast, and each second that passes reduces the popularity of a content, potentially increasing the long-term ROI of the next vote cast on it.
It's possible to chart the popularity that influences the DTC monetary distribution directly in the d.tube UI
This algorithm ensures there are always losers. The last upvoter never earns anything, also the person who upvoted at the highest popularity, and the one who downvoted at the lowest popularity would never receive any rewards for their vote. Just like the last upvoter and last downvoter wouldn't either. All the other ones in the middle may or may not receive anything, depending on how the voting and popularity evolved in time. The one with an obvious advantage, is the first voter who is always counted as 0 popularity. As long as the content stays at a positive popularity, every upvote will earn him rewards. Similarly, being the first downvoter on an overly-popular content could easily earn you 100% rewards on the next downvote that could be from a whale, earning you a fat bonus.
While Avalon doesn't technically have author rewards, the first-voter advantage is strong, and the author has the advantage of always being the first voter, so the author can still earn from his potentially original creations, he just needs to commit some voting power on his own contents to be able to publish.
ONE CHAIN <==> ONE APP
More scalable than shared blockchains
Another issue with generalistic blockchains like ETH/STEEM/EOS/TRX, which are currently hosting dozens of semi-popular web/mobile apps, is the reduced scalability of such shared models. Again, everything in a computer has a limit. For DPOS blockchains, 99%+ of the CPU load of a producing node will be to verify the signatures of the many transactions coming in every 3 seconds. And sadly this fact will not change with time. Even if we had a huge breakthrough on CPU speeds today, we would need to update the cryptographic standards for blockchains to keep them secure. This means it would NOT become easier to scale up the number of verifiable transactions per seconds.
Oh, but we are not there yet you're thinking? Or maybe you think that we'll all be rich if we reach the scalability limits so it doesn't really matter? WRONG
The limit is the number of signature verifications the most expensive CPU on the planet can do. Most blockchains use the secp256k1 curve, including Bitcoin, Ethereum, Steem and now Avalon. It was originally chosen for Bitcoin by Satoshi Nakamoto probably because it's decently quick at verifying signatures, and seems to be backdoor-proof (or else someone is playing a very patient game). Maybe some other curves exist with faster signature verification speed, but it won't be improved many-fold, and will likely require much research, auditing, and time to get adopted considering the security implications.
In 2015 Graphene was created, and Bitshares was completely rewritten. This was able to achieve 100,000 transaction per second on a single machine, and decentralized global stress testing achieved 18,000 transactions per second on a distributed network.
So BitShares/STEEM and other DPOS graphene chains in production can validate at most 18000 txs/sec, so about 1.5 billion transactions per day. EOS, Tendermint, Avalon, LIBRA or any other DPOS blockchain can achieve similar speeds, because there's no planet-killing proof-of-works, and thanks to the leader-based/democratic system that reduces the number of nodes taking part in the consensus.
As a comparison, there are about 4 billion likes per day on instagram, so you can probably double that with the actual uploads, stories and comments, password changes, etc. The load is also likely unstable through the day, probably some hours will go twice as fast as the average. You wouldn't be able to fit Instagram in a blockchain, ever, even with the most scalable blockchain tech on the world's best hardware. You'd need like a dozen of those chains. And instagram is still a growing platform, not as big as Facebook, or YouTube.
So, splitting this limit between many popular apps? Madness! Maybe it's still working right now, but when many different apps reach millions of daily active users plus bots, it won't fit anymore.
Serious projects with a big user base will need to rethink the shared blockchain models like Ethereum, EOS, TRX, etc because the fees in gas or necessary stake required to transact will skyrocket, and the victims will be the hordes of minnows at the bottom of the distribution spectrum.
If we can't run a full instagram on a DPOS blockchain, there is absolutely no point trying to run medium+reddit+insta+fb+yt+wechat+vk+tinder on one. Being able to run half an instagram is already pretty good and probably enough to actually onboard a fair share of the planet. But if we multiply the load by the number of different app concepts available, then it's never gonna scale.
DTube chain is meant for the DTube UI only. Please do not build something unrelated to video connecting to our chain, we would actively do what we can to prevent you from growing. We want this chain to be for video contents only, and the JSON format of the contents should always follow the one used by d.tube.
If you are interested in avalon tech for your project isn't about video, it's strongly suggested to fork the blockchain code and run your own avalon chain with a different origin id, instead of trying to connect your project to dtube's mainnet. If you still want to do it, chain leaders would be forced to actively combat your project as we would consider it as useless noise inside our dedicated blockchain.
Focused governance
Another issue of sharing a blockchain, is the issues coming up with the governance of it. Tons of features enabled by avalon would be controversial to develop on STEEM, because they'd only benefit DTube, and maybe even hurt/break some other projects. At best they'd be put at the bottom of a todo list somewhere. Having a blockchain dedicated to a single project enables it to quickly push updates that are focused on a single product, not dozens of totally different projects.
Many blockchain projects are trying to make decentralized governance true, but this is absolutely not what I am interested in for DTube. Instead, in avalon the 'init' account, or 'master' account, has very strong permissions. In the DTC case, @dtube:
- will earn 10% fees from all the inflation
- will not have to burn DTCs to create accounts
- will be able to do certain types of transactions when others can't
- account creation (during steem exclusivity period)
- transfers (during IEO period)
- transfering voting power and bandwidth ressources (used for easier onboarding)
For example, for our IEO we will setup a mainnet where only @dtube is allowed to transfer funds or vote until the IEO completes and the airdrop happens. This is also what enabled us to create a 'steem-only' registration period on the public testnet for the first month. Only @dtube can create accounts, this way we can enforce a 1 month period where users can port their username for free, without imposters having a chance to steal usernames. Through the hard-forking mechanism, we can enable/disable these limitations and easily evolve the rules and permissions of the blockchain, for example opening monetary transfers at the end of our IEO, or opening account creation once the steem exclusivity ends.
Luckily, avalon is decentralized, and all these parameters (like the @dtube fees, and @dtube permissions) are easily hardforkable by the leaders. @dtube will however be a very strong leader in the chain, as we plan to use our vote to at least keep the #1 producing node for as long as we can.
We reserve the right to 'not follow' an hardfork. For example, it's obvious we wouldn't follow something like reducing our fees to 0% as it would financially endanger the project, and we would rather just continue our official fork on our own and plug d.tube domain and mobile app to it.
On the other end of the spectrum, if other leaders think @dtube is being tyranical one way or another, leaders will always have the option of declining the new hardforks and putting the system on hold, then @dtube will have an issue and will need to compromise or betray the trust of 1/3 of the stake holders, which could reveal costly.
The goal is to have a harmounious, enterprise-level decision making within the top leaders. We expect these leaders to be financially and emotionally connected with the project and act for good. @dtube is to be expected to be the main good actor for the chain, and any permission given to it should be granted with the goal of increasing the DTC marketcap, and nothing else. Leaders and @dtube should be able to keep cooperation high enough to keep the hard-forks focused on the actual issues, and flowing faster than other blockchain projects striving for a totally decentralized governance, a goal they are unlikely to ever achieve.
PERFECT IMBALANCE
A lot of hard-forking
Avalon is easily hard-forkable, and will get hard-forked often, on purpose. No replays will be needed for leaders/exchanges during these hard-forks, just pull the new hardfork code, and restart the node before the hard-fork planned time to stay on the main fork. Why is this so crucial? It's something about game theory.
I have no former proof for this, but I assume a social and financial game akin to the one played on steem since 2016 to be impossible to perfectly balance, even with a thourough dichotomical process. It's probably because of some psychological reason, or maybe just the fact that humans are naturally greedy. Or maybe it's just because of the sheer number of players. They can gang up together, try to counter each others, and find all sorts of creative ideas to earn more and exploit each other. In the end, the slightest change in the rules, can cause drastic gameplay changes. It's a real problem, luckily it's been faced by other people in the past.
Similarly to what popular and succesful massively multiplayer games have achieved, I plan to patch or suggest hard-forks for avalon's mainnet on a bi-monthly basis. The goal of this perfect imbalance concept, is to force players to re-discover their best strategy often. By introducing regular, small, and semi-controlled changes into this chaos, we can fake balance. This will require players to be more adaptative and aware of the changes. This prevents the game from becoming stale and boring for players, while staying fair.
Death to bots
Automators on the other side, will need to re-think their bots, go through the developement and testing phase again, on every new hard-fork. It will be an unfair cat-and-mouse game. Doing small and semi-random changes in frequent hard-forks will be a easy task for the dtube leaders, compared to the work load generated to maintain the bots. In the end, I hope their return on investment to be much lower compared to the bid-bots, up to a point where there will be no automation.
Imagine how different things would have been if SteemIt Inc acted strongly against bid-bots or other forms of automation when they started appearing? Imagine if hard-forks were frequent and they promised to fight bid-bots and their ilk? Who would be crazy enough to make a bid-bot apart from @berniesanders then?
I don't want you to earn DTCs unless you are human. The way you are going to prove you are human, is not by sending a selfie of you with your passport to a 3rd party private company located on the other side of the world. You will just need to adapt to the new rules published every two weeks, and your human brain will do it subconsciously by just playing the voting game and seeing the rewards coming.
All these concepts are aimed at directly improving d.tube, making it more resilient, and scale both technologically and economically. Having control over the full tech stack required to power our dapp will prevent issues like the one we had with the search engine, where we relied too heavily on a 3rd party tool, and that created a 6-months long bug that basically broke 1/3 of the UI.
While d.tube's UI can now totally run independently from any other entity, we kept everything we could working with STEEM, and the user is now able to transparently publish/vote/comment videos on 2 different chains with one click. This way we can keep on leveraging the generalistic good features of STEEM that our new chain doesn't focuses on doing, such as the dollar-pegged token, the author rewards/donation mechanism, the tribes/communities tokens, and simply the extra exposure d.tube users can get from other website (steemit.com, busy.org, partiko, steempeak, etc), which is larger than the number of people using d.tube directly.
The public testnet has been running pretty well for 3 weeks now, with 6000+ accounts registered, and already a dozen of independant nodes popping up and running for leaders. The majority of the videos are cross-posted on both chains and the daily video volume has slightly increased since the update, despite the added friction of the new 'double login' system and several UI bugs.
If you've read this article, I'm hoping to get some reactions from you in the comments section!
Some even more focused articles about avalon are going to pop on my blog in the following weeks, such as how to get a node running and running for leader/witness, so feel free to follow me to get more news and help me reach 10K followers ;)