# Const's X Articles, 2026: full text Five long-form X Articles by @const_reborn (Jacob Steeves). The X API v2 export in `const_reborn-tweets.jsonl` holds only the `x.com/i/article/...` link for each; this file holds the bodies. All five were fetched on 12 Sep 2026 via the fxtwitter API, which returns the article content as Draft.js blocks. Bodies are converted to Markdown with no edits to wording, spelling, or punctuation. Like counts and other metrics are fxtwitter's values at fetch time; they match the like counts in the JSONL export. | # | Date | Title | Likes | Words | Status | |---|---|---|---|---|---| | 1 | 11 Apr 2026 | Response to Covenant | 2,385 | 1,070 | Recovered | | 2 | 19 Jun 2026 | Root Reborn | 525 | 1,467 | Recovered | | 3 | 20 Jun 2026 | Rebirthing the most unnecessarily contentious update in Bittensor's history | 752 | 1,524 | Recovered | | 4 | 21 Jun 2026 | Decentralization and Bittensor's Road towards it. | 1,088 | 1,647 | Recovered | | 5 | 23 Jun 2026 | Subnets mine TAO; TAO mines Subnets | 562 | 1,427 | Recovered | ## Summary All five bodies were recovered in full. Together they are Jacob's own long-form record of two crises and one design shift in 2026. The 11 April piece ("Response to Covenant") is the apology and counter-attack after Samuel Dare's Covenant subnets collapsed: he says sorry to everyone who lost money, calls Bittensor "by far the most decentralized AI protocol in existence", and proposes locked stake ("time + stake") as a cryptographic, not legal, measure of a team's commitment. The three June pieces on Root Reborn (19, 20 and 23 June) argue that the root network is "a dead subnet" that mechanically sells 1,000 TAO a day of subnet alpha; they turn validators into competing allocators of that yield, rebut Yuma's critique point by point, and re-specify subnet emission as a linear term (root proportion times price) gated by a boolean (miner burn), so a subnet that fakes mining can be switched off. The 21 June piece ("Decentralization and Bittensor's Road towards it") is the honest admission the thesis leans on: Bittensor "is currently not a decentralized protocol in the way Bitcoin is", it has leaders and "I am that person", the choice was speed over decentralization, and full decentralization and ossification are expected "within the next year and a half" (about December 2027). For the thesis at `docs/thesis.md`, these articles supply the primary source for section 12 (the key he holds, the dated path, "iteration before ossification"), for section 9 (no pre-mine, "no account with Satoshi's token stack", permissionless access with "no ID card"), for the "1,000-year Intelligence Federation" close, and for weak points 13, 16 and 17 (Root Reborn caps and validator competition, the Covenant fallout, three emission models in sixteen months). One tension to note: on 11 April he calls Bittensor the most decentralized AI protocol in existence; on 21 June he says it is not decentralized in the way Bitcoin is. The thesis reconciles this the same way the June article does: decentralized in ownership, not yet in core direction. A hostile questioner may still quote both lines. --- ## 1. Response to Covenant - **Date:** 11 Apr 2026, 21:45 UTC - **Tweet:** https://x.com/const_reborn/status/2043083033969033610 - **Article:** https://x.com/i/article/2043056814497886208 - **Engagement:** 2,385 likes · 652 reposts · 190 quotes · 329 replies · 455 bookmarks · 559,784 views - **Cover image:** https://pbs.twimg.com/media/HFpzllHWMAA3Vmx.jpg - **Source method:** Recovered on the first attempt from the fxtwitter API (`https://api.fxtwitter.com/const_reborn/status/2043083033969033610`), field `tweet.article.content`. The API returns the article as Draft.js blocks; I converted them to Markdown. Bold and italic runs are preserved as the author set them. The article has a cover image (https://pbs.twimg.com/media/HFpzllHWMAA3Vmx.jpg) and no inline media. Spelling and punctuation are as posted (for example "Artifiical", "is is", "it's power"). ### Body On a human level, I want to say that what transpired over the last few days shook me deeply. Someone I considered a brother, Samuel Dare, took actions with the intention to cause maximum pain towards me and the protocol I have given my life to. In that process he hurt all of Tao holders, most notably the people who believed in his subnets and the vision we originally shared for creating them. He let everyone down who bought his token, and who trusted in him. He betrayed us all. I want to sincerely apologize to everybody who lost money, who lost faith, who lost hope and had a financial rug pulled out from underneath them. I never thought I would see so much darkness these past few years, from people whom I have done my best to help in every way possible. It seems those you help the most can hurt us the most, and I have no explanation for that other than pointing towards deep mythological stories that have stood the test of time, which tell us that sometimes those who wrestle deeply with God, fail. Bittensor was built to be resistant to the pitfalls of human error; greed, selfishness, and other sins. Bittensor is a machine that is permissionless by nature because, whilst we are ALL fatally flawed, we are also fatally required to participate in the future of AI. Just as we cannot escape AI, AI should not escape us. It was this deeper wisdom which, upon realizing that Artifiical Intelligence would one day take over the world, awakened in me the knowing that one person should not own it, and that humanity as a collective must wield it. I saw it's power, and how it **must** belong to the people. To make it truly human we also need to make it truly open, which means accepting all of the darkest qualities of human nature into its core. For this reason I can't regret trusting Samuel, who did something incredible and ground-breaking with Bittensor, and for Artificial Intelligence as a field. He proved - though intense effort - that we *can* compete with centralized, billion-dollar, fiat-funded labs using open-source, permissionless networks and ***own*** ***those models together***. The profundity of that innovation cannot be overstated. The protocol Bittensor, the Internet of AI, an economic layer woven so deeply into the material world, must expect that the darker qualities in individual nature will inevitably come to light. I am certain that we, as a collective, alongside the machine itself, as a result of these attacks, can become resistant to them. We must. I am not going to respond more to the unfounded accusation of my friend Samuel here (as I have responded to them in many other places on X), whom I believe is likely having a psychotic breakdown, caused by flying too close to the sun. What I want to address is how we will move forward, and recover from the genuine threat to our network which Samuel's departure demonstrated with crystal clarity. In this industry, which is built on freedom, we do not have the privilege or the desire to lean on governmental and legal contracts which classically hold individuals accountable. These are slow moving, manual, ultimately corrupting and certainly not sufficient for the fast-paced AI world that we are entering. In this fast paced world where we all are headed, it will no longer be humans running subnets or companies. It will be faceless agents building digital system at lightning speed. Legacy understandings of value and law will soon be obsoleted and thus the real protection for us all is not legal but cryptographic. The teams and the branding will matter less, and the pure mathematical rules we build will govern us. We have to find a solution that protects investors and is is built out of **and improved** by the core cryptographic primitives of our industry. This is a problem that has plagued us all for many years and has not been solved anywhere in Web3.0 Thus, we are tackling one of the hardest problems in crypto; how do you measure a team's commitment to what they are building - aside from incentive and ownership - in open systems? ***Locked Stake*** That solution, ironically, was one of the last things that Samuel worked on before leaving the Opentensor Foundation. I consider my real error that we didn't implement it sooner, or perhaps we could have saved Sam from himself. The solution comes from the need to introduce a new aspect of ownership: *commitment*, which Is measured as the time remaining before tokens can move; time + stake. If passed, we will introduce this dimension as a protocol level feature which will allow subnet owners to explicitly define themselves as those that have conviction in the long term value of their commodities. This new transparency will allow investors predictability and assurances against events like this, and also open up a new quality through which teams can compete and create trust. With regards to the original subnets 3, 39 and 81. Many are already in the process of organizing members of the mining community and potentially (hopefully) ex-covenant team members to continue the work that was done, the code is open source and the vision for what was being built there has not changed - nor ever owned by one person. Functionally, these subnets should not change, and the holders/miners will organize around them to revive them. How we accomplish both these goals will require more organizing on the ground and I plan to go through these designs next week during the open Thursday call which you may all join in the Bittensor discord to ask questions and give feedback. I also want to take a moment to celebrate, and put our stake in the ground that we are ***by far*** **the most decentralized AI protocol In existence.** Bittensor is often overcomplicated, because the core technology is so abstract, however, we are fundamentally an economic layer that anybody can participate in, owning, mining and training AI. We are not perfect, and we have a ways to go, but we are the only group building SOTA artificial intelligence that is pointed towards being truly co-owned. And for that, I want to thank everyone who recognizes the urgent need our world has for an open and transparent artificial intelligence network. Next time we train 1 trillion. - const --- ## 2. Root Reborn - **Date:** 19 Jun 2026, 19:07 UTC - **Last edited:** 19 Jun 2026, 19:40 UTC - **Tweet:** https://x.com/const_reborn/status/2068048024987656603 - **Article:** https://x.com/i/article/2068020410071871490 - **Engagement:** 525 likes · 125 reposts · 54 quotes · 90 replies · 151 bookmarks · 156,852 views - **Cover image:** https://pbs.twimg.com/media/HLMdUnIXUAADosu.jpg - **Source method:** Recovered on the first attempt from the fxtwitter API, field `tweet.article.content`. The author typed Markdown by hand inside the article: lines starting `## ` and `---` rules. I kept the words unchanged but rendered his `## ` lines as `####` so they sit under this file's own headings. No cover image, no inline media. The closing line of section 10 ends in "..." in the original. ### Body Here is my response to the public critique made by Yuma on the addition to the way root prop is allocated, allowing the behavior already taken by a variety of institutions, to behave in an organized and transparent way on-chain. #### 1. Claim: Validators have no choice but to direct where root yield flows. No opt out. Some validators already direct their capital flows, for instance Crucible, Mentat, TrustedStake basket tooling, of which this proposal is modeled. A validator does not even need to opt out, as the default IS opt-out. In addition, they have another way of doing this within the upgrade itself, by setting full weights on all subnets. Both attain the staker a risk-free and non-opinionated rate. Expressing an opinion is already the choice for a validator today which automatically sells all of your alpha yield to TAO. Opt-out is a statement, and you can still be a validator or stake to a validator that does that. --- #### 2. Claim: Creates a single escrow account, which adds structural risk. This is an egregious error. The escrow is not an account; the mock-coldkey is a placeholder owned by the chain, simply a technical choice which makes the accounting system elegant. It is not an account like yours or mine — it cannot be stolen, it cannot act like a honeypot, and it cannot unstake funds. --- #### 3. Claim: Redemptions triggers bank runs with no circuit breaker. Claims are just an unstake operation across the basket. Do alpha holders cause bank runs on Bittensor today when they claim their yield across their holdings? No circuit breakers are required today, and an abstraction over the same system does not beget one. --- #### 4. Claim: A validator's influence over the basket fades as the basket grows. You are not staking into a validators basket, nor buying into a set of tokens; you are attaching your TAO to a distributional vector (weights) which performs forward in time. Technically each staker's basket is distinct and based on when they joined and when they leave -- in other words basket's size and age are irrelevant to a consumer. What sticks is an validators historical performance (how the vector performed over time), but this is entirely the point of the upgrade. --- #### 5. Claim: Subnets must lobby for weights instead of building. Subnets on Bittensor already fight for 4–10k TAO worth of daily volume, which comes from investors, institutions, and individuals. Root reinvestment today would be an additional 1,000 TAO. The largest validator on Bittensor can direct 150 of that, and this is across a basket of multiple subnets. At best, a subnet just found an additional liquidity provider with a constant drip of support (more buyers); at worst, they lost a percentage of additional global investment — and their total remains unchanged (100% root sell pressure as it was before). --- #### 6. Claim: Validators are now under regulatory pressure where they were not before. What this proposal adds is optionality — including two explicit, validator-controlled ways to remain as passive as the day before. Root dividends do not arrive as TAO; they arrive as subnet alpha, and the protocol takes the action of selling that alpha for TAO on behalf of root stakers before they ever see it. The change gives you and validators the option to curate that if you and they want. --- #### 7. Claim: less sell pressure on subnets and tax basis For holders, this change provides additional choices and the opportunity for higher yields — alpha exposure can attain them a higher yield than a flat TAO emission, and they keep the option or take neither. For taxation, it is the inverse: root yield is now on-demand so holder controls when realization happens, and can thus defer. The old model realized for them constantly. Fewer taxable events and deferral optionality is simpler for the holder, not more complex. On reduced sell pressure raising miner rewards, its hard for me to understand how this could even be put into an article. More miner rewards is clearly a feature. Less automatic dumping of root TAO, recycled into the ecosystem, is net supportive of the whole network — miners included. Absurd that this was listed as a concern. On subnet stakers getting "lower emissions as that gets divided with root stakers". Well exactly, we are getting more for root stakers and this admission proves it -- thats the whole point. --- #### 8. Claim: Structural three way slippage Equivalent redistributive vehicles like Crucible and TrustedStake already pay the exact same slippage and their products strongly out perform the market. This is ONTOP of the fact that they pay staking transaction fees on behalf of the participant (this proposal does not include those). --- #### 9. Claim: Structural: chain scale / unbounded execution Thankfully the wonderful Vune and Greg already solved these problems last year when we built the lazy accounting systems (hidden from a users view) that allow the chain to distribute tokens to all stakers on a validator in an O(1) single operation. The critique from Yuma simply doesn't know the chain code. On subnet de-registration we pay the same overall computational cost as we did before, it's an iteration over accounts -- but this is not changed by this PR. --- #### 10. Claim: LIBOR analogy LIBOR was an opaque, non competitive, rate setting system in which institutions could make choices without any skin in the game. Validators are under the opposite regime: it is transparent, competitive, and with skin in the game: bad choices effect them directly. Transactional markets are healthier than rate-setting cabals — is an argument **for** this proposal. We are not in the root network of 2024, of which nobody benefited more through corruption and egregiously corrupt root weights than ... --- #### 11. Claim: Validator insider dealing. One of the main concerns with this PR is that validators collude with subnet owners as they did in 2024. But we are not in 2024, we have a dynamic market into which we can evaluate performance. There is a non-subjective truth against which validators must guide their choices and that is the performance of the subnets under the market today. As explained above, a single validator's root-derived buying pressure is extremely small relative to the overall size of the dynamic market. Any validator attempting to artificially boost yields would need to create and maintain upward price action themselves i.e. real buying themselves. Those alpha holders they have now allocated can front-run and sell into the strength simply by unstaking from the bad-acting validator, hurting its network dominance and making the scheme increasingly expensive to sustain. The exploiter effectively has to prop up the market to maintain the yield, which creates a cost. If the economics stop making sense or the price support fails, stake naturally migrates elsewhere in search of higher yields. In practice, the attack becomes self-limiting because maintaining it requires continuous capital commitment and market risk rather than being a free extraction mechanism. Such an attack is theoretically possible, but the incentives are further eroded when you consider that validators are an active, non-anonymous class of organizations. Their weights are transparent, their teams are known, and their behavior can be publicly scrutinized (unlike anonymous investment today), and the system naturally checks bad behavior by undermining their yield. This is a far cry from the days of the original root network, where validators risked nothing (could not be checked) and were collectively directing 100% of Bittensor's TAO inflation with no cost to their stakers. --- #### Conclusion The risk with this change for validators is not regulatory, technical or otherwise: They can opt out of this system and continue doing what they are already doing today. Nothing changes for their stakers except for the need to call a claim operation which defers their tax event. The risk with this change is not any of those written above, it is competition that young smart teams will enter the root network and bring optimized staking products to the ecosystem which help direct Bittensor, bring better returns for stakers, prove the investment case for Bittensor, and in so doing undermine the hegemonic control of large entrenched incumbents. What investors in Bittensor needs to understand is that we are in a fight to provide the world's most sophisticated alignment machine ever invented — one capable of out-competing the AI giants by the raw horsepower of economic markets. Those who worry about process are partially correct: we can always do better with messaging and public relations and rollouts. We move fast because the world moves fast and what matters here is succeeding for humanity not the comfort of passive investors. Cyberpunks write code. Institutionals apparently write AI slop articles on twitter. — @const_reborn --- ## 3. Rebirthing the most unnecessarily contentious update in Bittensor's history - **Date:** 20 Jun 2026, 00:59 UTC - **Tweet:** https://x.com/const_reborn/status/2068136709439517112 - **Article:** https://x.com/i/article/2068089653396185088 - **Engagement:** 752 likes · 217 reposts · 70 quotes · 94 replies · 227 bookmarks · 179,900 views - **Cover image:** https://pbs.twimg.com/media/HLNUtUvXkAA1BLw.jpg - **Source method:** Recovered on the first attempt from the fxtwitter API, field `tweet.article.content`. Section headings were `header-two` blocks; I rendered them as `###`. Eight inline images are linked to their pbs.twimg.com originals; two carry the author's captions, shown in italics under the image. The trailing space in the title is in the original. The final line is a bare link to the pull request, as posted. ### Body > Firstly; this change is optional. If you do nothing, you can keep staking as usual and claim your TAO rewards when you choose. The wallet experience will remain simple: stake, earn, claim. > ***But if you are interested, if you want to help make Bittensor work as a machine, and I mean a machine that goes prrrr, like a Jaguar F1, read on:*** ### TAO is not Solana, and it is not Ethereum ![image](https://pbs.twimg.com/media/HLNV4ngXkAAbwJY.jpg) Most crypto assets are bets that transaction fees and network usage will sustain hundreds of billions of dollars of market cap. TAO is different. TAO is a productive utility token that mints a return for the people who hold it. TAO achieves this by incentivizing its subnets: the chain gives liquidity to 128 competitively pitted companies, each of which converts **commodity mining networks** into revenue. Bittensor gives them that investment, a stack to build on, and a network of miners that make their projects function. In exchange, TAO itself makes a return by getting a piece of those running companies: a proportion of their alpha tokens, minted block by block, and then sold as a stream of sell pressure back to TAO holders over time. This is called “root proportion”: the proportion owed to TAO. Today, Bittensor produces 1000 TAO of return on those projects for the 3600 TAO of global investment added as liquidity to those subnets. It sells 3600 TAO and gets 1000 back. ### Yield leakage Bittensor is not lacking yield (it is actually one of the highest revenue-to-market-cap crypto projects in existence), but that's a low bar, and we could do A LOT better. > Note: 65% of the remaining 2600 TAO spend is paired with subnets i.e locked in liquidity pools (not given out as sell pressure) or paid back when a subnet deregisters. The network receives alpha from subnets, but instead of letting capital compound intelligently, it sells that alpha mechanically. This creates immediate sell pressure, damages price discovery for subnets, and turns potentially valuable ownership into short-term TAO flow. The problem is not that subnets pay Root. The problem is that Root has no intelligence. It burns valuable ownership in early-stage projects and disrupts the market signal for the underlying assets in exactly the wrong way: downward. But lets invert this also: we are paying TAO on root yield, what are getting from that passive activity? The 1000 TAO per day could also be used to organize the network intelligently. We can do better. ### The dull network ![image](https://pbs.twimg.com/media/HLNWdS8W8AAAS5D.jpg) "Root" is subnet 0: the null network, TAO’s own network. It has no miners, no separate token, and no company behind it. It is where TAO holders already delegate stake so validators can express weights across the network. Today Root is mostly passive. It is powerful, but currently useless. Completely nerfed since the upgrade to dynamic TAO and given little control or sway in the network. It controls one of the most important allocation surfaces in Bittensor, but it is not yet a real competitive mechanism. It is, effectively, a dead subnet. ### Bittensor as an incentive machine. ![image](https://pbs.twimg.com/media/HLNa1BlW4AAnrHP.jpg) The entire purpose of the Root Reborn upgrade is to ignite the root network. Instead of merely selling subnet yield, having its validators compete to allocate, reinvest, and compound that yield. Root then becomes a meta-optimization layer over the entire subnet economy. Normal subnets optimize miners. Root optimized subnets baskets. And we win both ways, higher yield and more intelligent allocation. ### A meta-incentive layer for TAO ![Root validators and their yields today.](https://pbs.twimg.com/media/HLNrGQ9WkAAmfs4.jpg) *Root validators and their yields today.* This network could and should be so much more: a meta-incentive layer: Validators are smart, they are aligned, and they are technically aware and care about the Bittensor network, they should become its first transparent allocators. TAO holders can choose validators based on performance, where there exists an understandable scoreboard already: who produced the most TAO-denominated return for their stakers over time? Instead of optimizing weights over miners to produce commodities, there is a clear and symmetric inverted mechanism through which we elicit validators to optimize weights over subnets to produce yield. ### Mining Root, in a sense. ![image](https://pbs.twimg.com/media/HLNxuGnW8AEBqIN.jpg) People want this, and the demand has already birthed this exact mechanism from companies like Trusted Stake, Crucible, and Mentat Minds, young smart teams looking for an edge, that have built the same product three times: an off-chain system where they take the yield you get from sold alpha tokens and reinvest it into the very system that sold it. Building it off chain is not ideal though, it **is** far less efficient what the current update will facilitate. It is manual, running once per day or once per week. It costs additional transaction fees, and recently led to investors losing thousands of TAO through lack of on-chain safety mechanisms. But the structure is there already, it is tried and tested, and it has been shown to outperform the market. But these are not standardized products. They follow different rules, they are hard to compare, and they are not naturally competitive with each other. Making this layer standardized, competitive, and transparent across the whole network will certainly improve this service. ### How it will work for you You look at your wallet and pick a performant validator, or a set of validators. You stake to all of them, or none of them, or the passive ones that take no view, or the active managers helping the network form a loose consensus about what Bittensor holders want to support and what they do not. Validators can now earn their reputation and trust based on the yield they provide over time. They will garner support from passive holders of TAO, improving the structure and stability of solid, convicted, successful subnets within the ecosystem. While optimizing yield through reinvestment, they will participate more actively in supporting good subnets that deserve a more stable form of capital investment from the network. Theoretically, and in practice, this will provide an immense boon for the good subnets on Bittensor and enable the validator class to direct emissions to subnets providing value. ### What keeps it honest? The pre-dTAO Root network was corrupt. But unlike Root in pre-2025 Bittensor, there is a check on this system: raw yield over time. We are not the same ecosystem as we were back then. Badly performing baskets dip below their competitors and lose stake. Downside and skin in the game are here, visibly. Better allocators for TAO holders rise to the top. New teams enter the validator class. More active, intelligent, hard-working groups make the network run. It is a transparent, contestable competition, where the scoreboard is how much TAO you made for your stakers. Insider dealing is visible in a competitive market, and spread across many validators. It is hard for any one of them to move the scale in a meaningful way, and any unfair support they do provide exposes both them and the subnets. TAO holders can easily deallocate from both alpha holdings and the validator. ### Flywheels all the way down ![image](https://pbs.twimg.com/media/HLNZWvXWEAAT0L1.jpg) When you stake into validators that set their weights well, and those validators collectively reach consensus on good choices, the network leaks less value because TAO gets reinvested into good projects instead of dumped by shit ones. TAO earns a higher risk-free rate, so your base yield goes up. It attracts more fundamental value, because a productive asset with a real, visible rate is a better thing to own. It pays itself back over time, as capital compounds into the subnets that actually perform. The side effects compound the story. More TAO flows into subnets, which means more investment for the deserving teams building on Bittensor. That lifts the sum of subnet prices, which is what lets the network sustain high yields in dollar terms. ### Lets not force sell subnets! ![Sum of subnets since November TaoFlow update.](https://pbs.twimg.com/media/HLNyjsxXwAEUuAh.jpg) *Sum of subnets since November TaoFlow update.* There is also a quieter benefit that matters enormously for price. Today, Root yield, or Root sell pressure, is realized continuously: a stream of immediate taxable events and a stream of automatic sell pressure. Going forward, the chain will collect Root yield and not release it until a staker chooses to claim. That single change converts immediate taxable events into deferred ones, and it could remove up to 33% of Root-induced sell pressure per year from immediate taxes. ### Conclusion ![image](https://pbs.twimg.com/media/HLNZQAfXkAA-zSj.jpg) We are turning the root network into the engine that maximizes the value of holding TAO. In the process we are making TAO a much more attractive investment and bringing to Bittensor a new class of intelligent actors. I believe that if we continue to move in this direction we can and will merge the power of Bitcoin with the adaptability of AI we know exists, and succeed on the mission of building an open intelligence network. Adelante comandantes, @const_reborn P.S. This is the first of a number of articles OTF and I will be releasing that map out how and what we intend to bring to this ecosystem in the near future. https://github.com/opentensor/subtensor/pull/2759 --- ## 4. Decentralization and Bittensor's Road towards it. - **Date:** 21 Jun 2026, 18:17 UTC - **Tweet:** https://x.com/const_reborn/status/2068760333787808043 - **Article:** https://x.com/i/article/2068728221114241025 - **Engagement:** 1,088 likes · 306 reposts · 120 quotes · 124 replies · 282 bookmarks · 257,816 views - **Cover image:** https://pbs.twimg.com/media/HLW1lAZXIAANmd5.jpg - **Source method:** Recovered on the first attempt from the fxtwitter API, field `tweet.article.content`. Section headings were `header-two` blocks; I rendered them as `###`. The three bullet items were `unordered-list-item` blocks. No cover image, no inline media. ### Body Bittensor is currently not a decentralized protocol in the way Bitcoin is. It can be, and it will be, but Bittensor's vision is the unification of two technologies: the fast-paced, mutable substrate of AI plus incentive, and the ethically rooted, immovable substrate of Bitcoin. Bitcoin was built to be resistant to a nation-state takeover. It needed to be that because there is nothing more threatening to the "powers that be" than a competitive and fair system of money. Bitcoin was born into a fully fledged financial system of near-total control, and it was essential to be fully decentralized and anonymous from day 1. AI is equally as important for the future of humanity. However, it is not under the same level of attack today, as it is an emerging tech. Neither culturally nor legislatively have we remotely begun to figure out our ability to own, or have sway over, the future of intelligence. This beast is only beginning to rear its head. Mythos and the closing off of intelligence are just the beginning of a long path. Bitcoin was the end result of one. ### Where we are at: Bittensor is built on one of the most profound, well-built technologies ever created: Substrate, Polkadot tooling for the creation of blockchains, which is upgradable, connectable, hard-packed with an SDK and cryptography to make it resistant to any group while remaining mutable, extremely fast-running, and decentralized. Bittensor, at its core economic incentive layer, is currently not decentralized. We have the potential technology underneath us, and it is one of our main goals. But we have not yet chosen to flip the ultimate switch and make it run like the nation-state-resistant bunker that it will one day become. Instead, Bittensor has leaders. There are people at its helm: me, two other engineers, and a very tight-knit group of absolute geniuses, holders, and believers who have been here fighting since the beginning to make sure that this machine is aligned, working, purring with efficiency, self-sustainable, and the best place to build. There is no point in decentralizing a system that doesn't yet deserve decentralization, and in the fast-moving field of AI, we have continually tacked Bittensor in the direction of better and better and better. From Kusanagi to Nakamoto to Finney to Dynamic to Flow. Anyone with eyes can see and appreciate the 0-to-1 innovations that Bittensor has gone through, and needed to go through, to reach further and further along the trajectory of a decentralized AI network that can actually make a difference. We have decided, and I maintain that this is the correct decision, to move fast and iterate, update, and develop quickly, at the cost of remaining centralized, rather than sacrificing quality and moving slowly and "democratically." I have noticed there are three camps of people/groups in Bittensor who have something to say about the updates we push: - People who are TAO holders and believe in the vision, but who are not technically proficient enough to understand exactly what is going on and want more communication For these people, my commitment going forward is to release a long explainer article alongside each PR update, such as the one I released yesterday, for the average TAO holder/casual non-technical investor. I will address the problems in the network's existing state that the update is improving. Any questions or hesitations that people may have will be thoroughly addressed. I have heard you loud and clear, and I am committed to explaining Bittensor and every update we push going forward in layman's terms so you will all be excited. - People who are technically proficient and understand Bittensor and the PR in code form These people are almost exclusively supportive of the updates we make. These are the subnet owners, miners, and builders. Their feedback is almost universally "this is how we can improve" or "this is where this has flaws," rather than "let's not move forward." I already take these people's feedback to heart and use it to strengthen our ideas. You know who you are, and thank you. - Grifters, scammers, people with an axe to grind, and those who want to use the concept of "decentralization" to prevent us from pushing updates that will prevent their continued grift. I obviously won't be listening to these people. Whilst unfortunately they can oftentimes be the loudest voices, I hope the articles and additional communications and explanations I am committed to providing will help quieten the power of these voices. ### How we are decentralized: Bittensor is also already decentralized. People may critique where we are not, but let me show you where we have achieved the important things upfront, the things that matter the most and cannot be done later. Bittensor was not pre-mined, and it has been online for 5+ years, distributing its ownership layer outwards to people, real, honest, and brilliant people who have worked and brought innovation for access, rather than being gifted it or having it bureaucratically distributed. We are the only crypto-AI project that did it right, as far as I can tell. We have been rewarded by that nascent, authentic community, plus more raw talent than you will find anywhere else in our field. You do not need permission to build a subnet, mine a subnet, study, or use our technology. You do not and will never need an ID card to access AI built on Bittensor. At an economic level, Bittensor has one of the best allocations ever achieved. Better even than Bitcoin itself, as there is no account with Satoshi's token stack on Bittensor. Bittensor is also a live ecosystem, with 128 subnet teams, unique CEOs, developer teams, and communities. We have more than 20 core validator teams and many other smaller ones. Bittensor is decentralized where it matters: its ownership. ### How we are not decentralized, and why? Where Bittensor is not decentralized is with core direction. Bitcoin was no different. For the first 2.5 years of Bitcoin's life, it was Satoshi who called the shots and wrote the tech. The core frameworks, the core tooling, the position of the parties, everything was one group, or one man, who made it that way. I am that person for Bittensor. Behind me are the people I trust the most, who sign off on core updates that need to go through and who say no if there is a fault that needs to be corrected. And then there are you, the people in this ecosystem who loudly give feedback and tell us where we are making mistakes. We need both. I integrate myself into every conversation, every chat, and every thread. I believe many of you know how much I care about how this vision lands with people. ### What the roadmap is in terms of future updates, and the ultimate goal of true decentralization Let's talk about where we are going. We currently have an incentive machine that has pulled unbelievable talent from the darkness and applied it together in a permissionless system to build SOTA intelligence. But there are some things that still need building. Last week I outlined one of them: we need to bring validators back into the competitive game and elicit them to help organize and optimize the investment case for TAO. This update translates the Root subnet of Bittensor into a flywheel, which drives more economic utility into TAO. We will continue to develop and update validators' expansion as active, competitive players in this ecosystem. Next, before we can make the emission allocation on Bittensor purely market-based, we also need to open both sides of its liquidity pools to investment: pool borrowing, or, colloquially, shorting. This will make the market symmetrical, so that attackers cannot so easily manipulate the on-chain signals that drive the growth of the Bittensor network. This is essential for the hyper-optimization of Bittensor's internal market and the production of trustless swarm wisdom to be the final arbiter of Bittensor's incentives. We need to turn on conviction, giving alpha-token holders a power structure based on a team's commitment to the future success of their token and/or a competitor. This is to say, we need to bring alpha-token holders' rights. Additionally, expect updates to TaoFlow and its derivatives in the coming weeks, with more fine-tuning of the emission allocation algorithms in relation to how Bittensor programmatically allocates its inflation. As any subnet owner knows, incentive mechanisms are an art that you need to get right through fine-tuning. DTAO is no different. Finally, we are going to take a more active role in excluding teams from the ecosystem that are not actively participating in adding value to the network and instead are exclusively extracting. All of these updates will dramatically improve the efficiency of the Bittensor network and reduce global leakage of value, driving more and more value to the builders on Bittensor who share our vision, and less towards those who seek to undermine it. ### Path to decentralization. We will know when it is time. I expect it will be within the next year and a half. Making these three pillars work together will be the completion of the core mechanism that makes Bittensor work. Aligned incentives, optimized value, and true ownership rights. A network that runs faster than any centralized company, whilst giving everyone a piece of it. Bittensor is not complete until thoroughly decentralized, and decentralization is and will remain the crowning jewel of what we are building. It is not complete until we can root this entire throbbing orb of potential directly into the substrate we built on from the get-go. Its final ossification. Its total decentralization. Its programmatic immutability is, and will always remain, the final goal. This will come soon, when we throw the keys away on this spaceship and let it be what it was always meant to be. 1,000-year Intelligence Federation. I cannot wait. Blue Skies, @const_reborn --- ## 5. Subnets mine TAO; TAO mines Subnets - **Date:** 23 Jun 2026, 01:21 UTC - **Tweet:** https://x.com/const_reborn/status/2069229442522599847 - **Article:** https://x.com/i/article/2069161473633615872 - **Engagement:** 562 likes · 117 reposts · 46 quotes · 49 replies · 135 bookmarks · 112,026 views - **Cover image:** https://pbs.twimg.com/media/HLcjc6qawAA_pbG.jpg - **Source method:** Recovered on the first attempt from the fxtwitter API, field `tweet.article.content`. The seven equations were `LATEX` atomic blocks; I rendered each as a `$$` display-math block with the LaTeX source unchanged. The GitHub link was a `LINK` entity whose visible text is the URL itself. No cover image, no inline media. ### Body $$ e_i \;\propto\; \underbrace{\rho_i \times \bar{p}_i}_{\text{linear (maximize)}} \times \underbrace{(1 - b_i)}_{\text{boolean gate}} $$ > Disclaimer: this upgrade only effects subnet owners and dynamic TAO traders, it does not have any effect on TAO/Alpha holders, wallet interfaces etc -- it is purely an upgrade to Bittensor's core incentive system. In all optimization, what we measure, we can manage. Managing a subnet on Bittensor is no different: the first thing you need to ask is what are we measuring? Not in a handwavy manner -- in as clear and concise a way as possible: the metric, the benchmark, the number. Here is an example: on Affine, we measure performance against 5 RL environments by a language model which has the Qwen architecture, tokenizer, and 33B parameters. Here is another: on Lium, we measure how many GPUs you have brought to the network which pass a reliability filter. And another: on Chutes, they measure how many inferences you can sustain while running a trusted execution environment. After years of building some of the first ever digital commodity systems, we've found they almost all have the same structure, composed of two things: (1) a linear term -- "more of some quality" -- and (2) a binary term, a boolean flag, pass or fail. $$ \text{reward} \;=\; \underbrace{Q}_{\substack{\text{linear}\\\text{maximize} \;\uparrow}}\;\times\;\underbrace{\mathbb{1}\!\left[\text{requirement met}\right]}_{\substack{\text{boolean}\\\text{pass / fail}}} $$ On Affine those terms are performance and architecture; on Lium, quantity and minimum quality; on Chutes, bandwidth and TEE proof. The structure is good for a clean reason: we want more of #1, but we can't be sure we're actually maximizing it unless #2 holds. The two terms do different jobs. Some properties we don't want to optimize but do need to require -- a GPU needs at least a 50Mb connection, but we don't want to pay more for more bandwidth, so it can't live in the linear term; it has to be a binary requirement. In practice, miners push right up against the boolean while maximizing the linear term -- "pay the minimum bandwidth you can while bringing the most GPUs." That tension is the structure of a good incentive system. $$ \begin{array}{l|l|l} \textbf{Subnet} & \textbf{Linear (maximize} \uparrow\textbf{)} & \textbf{Boolean (pass/fail)} \\ \hline \text{Affine} & \text{performance, 5 RL envs} & \text{Qwen arch, tokenizer, 33B} \\ \text{Lium} & \text{number of GPUs} & \text{reliability filter} \\ \text{Chutes} & \text{inference bandwidth} & \text{TEE proof} \\ \text{Bittensor} & \rho \times \bar{p} & 1 - b \ (\text{burn key}) \\ \end{array} $$ Bittensor is also a subnet. Designed from the ground up to be recursive and fractal, subnets are obviously subnets -- and one day they will have subnets of their own. A subnet is just an optimization algorithm whose definition of reward drives miners to behave a certain way, and Bittensor has the same shape, with its primary term being emission: the thing you get for maximizing the algorithm. So, like any good subnet, Bittensor must have its own definition of success: (1) a linear term we push, and (2) a boolean term we push against. We upgraded the chain today to make this explicit. **The linear term is root_prop × price.** This is the quality we maximize -- the per-block value returned to root. $$ \text{emission\_share} \;\propto\; \text{root\_prop} \times \text{price} \times (1 - \text{miner\_burn}) $$ Price is the most natural and intuitive metric for the chain to optimize: it is directly proportional to the value attained for the dilution of TAO. It is also the metric dynamic-TAO investors care about most, because it underpins the quality of what mining on Bittensor can produce and reflects healthy internal markets. Concretely, price here is the subnet's exponential moving price -- the smoothed pool price, not the spot tick. Combining price with root proportion tunes this further. root_prop -- ***tao_weight / (tao_weight + alpha_issuance)*** -- is the per-block share of a subnet's dilution that is sold back to TAO holders. Multiplying it through ties the core optimization explicitly to the proportion of value that returns to TAO, and gives the system a clear north star. It has a second, deliberate effect worth stating plainly: because root_prop falls as a subnet's alpha issuance grows, emission naturally eases toward newer subnets and decays as a subnet matures. New entrants get an easier on-ramp; incumbents must keep earning their share on price. That is intended, not incidental. $$ \rho_i \;=\; \frac{T_{\text{root}}\, w}{T_{\text{root}}\, w + \alpha_i} \qquad \begin{aligned} &T_{\text{root}}:\ \text{root TAO stake (network-wide)}\\ &w:\ \text{tao weight (network-wide)}\\ &\alpha_i:\ \text{subnet } i \text{ alpha issuance} \end{aligned} $$ **The boolean term is** *(1 − miner_burn)***, and it is operated by validators.** This is the requirement, not the optimization — the flag we push against. miner_burn is the proportion of a subnet's miner emission that is withheld from miners (routed to the owner/burn key and burned) in a given tempo. The key change is that burning miner emission now costs *chain* emission, not just internal subnet emission: send your miner rewards to the burn key and your subnet's share of the network falls in lock step. $$ b_i \;=\; \frac{\displaystyle\sum_{k \in \mathcal{O}_i} I_{i,k}}{\displaystyle\sum_{k} I_{i,k}} \qquad \begin{aligned} &I_{i,k}:\ \text{miner incentive to hotkey } k \text{ on subnet } i\\ &\mathcal{O}_i:\ \text{owner / burn-key hotkeys} \end{aligned} $$ I want to be precise about why this is a boolean and not a second linear term. We are not trying to measure "good mining" on-chain -- that's a fool's errand and the source of endless gaming. Instead we hand the judgment to the people best positioned to make it: validators. Validators are a distributed set, and in practice they operate this lever the way any binary is operated -- burn everything or nothing. A subnet that runs a real, active mining mechanism keeps its emission; one that doesn't can be switched off. This judgment will be driven in conjunction with the new oversight tooling from the triumvirate, which exists to cut emission on fake-mining subnets in the cases where the validator set is itself ineffectual or inactive. The chain provides the lever; validators and oversight provide the verdict. $$ (1 - b_i) \;\approx\; \begin{cases} 1, & \text{active mining (burn nothing)} \;\Rightarrow\; \text{full emission}\\[4pt] 0, & \text{burned off} \;\Rightarrow\; e_i = 0 \end{cases} $$ The same lever has two honest users, and the chain is deliberately agnostic about which one pulls it. A team can manage its own emission -- choosing to burn rather than over-dilute -- but if it does, it pays in immediate emission. A validator (or the triumvirate) can burn a subnet that is faking its mechanism. Both move (1 − miner_burn) the same direction, and the desired outcome — less active mining means less emission -- is identical regardless of who acted. This is also our commitment to the mining side of Bittensor made explicit: mining subnets should no longer sit at a disadvantage to non-mining ones. If you don't have miner emission, the chain follows suit. We're aware this invites teams to try to sidestep running ignited mechanisms, and that the burn term raises the oversight bar. That's expected -- and it's why our answer is governance and validation, not an on-chain mining detector we'd only have to keep patching. This upgrade also fixes a structural flaw in the previous mechanism. TaoFlow measured subnet performance on a moving average of netflow -- how much TAO left a pool over a window. But moving averages have memory, and they forget. That created a clean arbitrage: buy one of your subnets, collect emission, sell out into a negative EMA, rotate to the next, and come back once the first pool had "forgotten." Cycling emission between your own subnets was a strategy. Price-based emission has no such cycle. Because of the V2 pools we use, it is fully symmetric: a purchase and a sale move the emission vector in equal and opposite proportion, regardless of when you make them. And the boolean inherits the same property -- miner_burn is computed fresh each tempo, with no history. A subnet that stops burning recovers its full emission the very next tempo; one that starts burning loses it immediately. There is nothing to wait out and nothing to arbitrage on either term. As an important note to end on: Bittensor itself is a subnet, and as all subnet owners know: the relentless tide of miners against our mechanisms is both our greatest strength and a continued fight. We would not be Bittensor if we didn't have a commitment to consistently and continually tune and improve these systems. As ever, more to come. Best, @const_reborn For those who want to read more, the code is here: https://github.com/opentensor/subtensor/releases/tag/v3.4.6-421. --- ## Method log 1. Located the five tweets in `const_reborn-tweets.jsonl` by filtering `entities.urls[].expanded_url` for `x.com/i/article/`. Exactly five non-retweet entries matched, and their dates and like counts match the five flagged in `docs/research/const-tweet-themes.md`. 2. Queried `https://api.fxtwitter.com/const_reborn/status/` for each. All five returned HTTP 200 with a populated `tweet.article` object (`title`, `preview_text`, `cover_media`, `content.blocks`, `content.entityMap`, `media_entities`). 3. Converted the Draft.js blocks to Markdown: `unstyled` to paragraphs, `header-two` to `###`, `blockquote` to `>`, `unordered-list-item` to `-`, `MEDIA` atomics to image links with captions, `LATEX` atomics to `$$` blocks. Inline `Bold`/`Italic` ranges were applied using UTF-16 offsets, as Draft.js requires. 4. The fallback methods listed in the assignment (WebFetch of the article URL, headless browser, nitter mirrors, web search for reposts) were not needed and were not run.