Hello X, I wanted to share what I have been working on for the past few months: This Q2 report covers Aave protocols full on-chain financial activity for the first half of 2026. The report covers over 1.3 Million transactions over 2,000 addresses and contracts. The main statements (only Ethereum is covered) are: - Full Aave protocol income statement - Full Aave protocol balance sheet - Aave DAO Treasury cashflow - Aave DAO balance sheet The financials aren't built by hand. I created a data pipeline that extracts blockchain txns from RPC nodes, classifies accounts based on token movement and log traces, then transforms the classifications into journal entries. Then I built some custom agent skills to connect the classification to accounting standards.

I think @bankrbot might be heavily undervalued. Crypto AI tokens are worth $23.2B combined, but very few projects make money. This chart plots each project's market cap against a year of revenue from outside customers. The cheapest relative to their revenue: 📺 Bankr: 1.4x ⚡ Aethir: 2.5x 🔲 io.net: 5.5x 🌱 Grass: 15x The most expensive: 🔍 The Graph: 1,178x ♾️ ICP: 736x 🎨 Render: 434x Five tokens worth a combined $2.6B disclose no revenue at all: Pearl, Akedo, Quack AI, Unibase and Kite. Unibase has settled $28 of payments in total. I could be wrong about buybacks, where revenue comes from, and other aspects but from a pure mcap/revenue multiple it looks like the base:0x22af33fe49fd1fa80c7149773dde5890d3c76f3b is the cheapest on the market.

How PaperTrade got started: The story of how a high school gamer turned trader raised $55M in fees to holders. Founder @izebel_eth started in high school as a gamer who traded loot boxes and flipped Team Fortress 2 hats, clearing his inventory every day for small profits. Later he moved into options trading, and then into high-frequency options market making as a career. He worked in automated market making at Morgan Stanley and as a quantitative research analyst at Millennium. Between about 2015 and 2019 he was making markets for the early WallStreetBets traders. Soon he wanted to trade crypto, but Millennium’s compliance rules blocked it. He talked to Hudson River Trading and Jane Street about moving firms, then left the industry instead and went back to trading his own money full time. From 2017 to 2021 he spent most of his time in a private Discord, trading and talking under the handle izebel_eth. One of the anonymous members in that Discord was the founder of a company Standard Crypto had invested in. When people at the firm asked that founder for reliable contacts in crypto, he gave them Jez’s Discord username, and that was how Jez got brought in. He joined as a Venture Partner, which let him keep trading his own book. While he was there he sourced the firm’s investment in Blast and wrote the public thesis for it on his Substack. He also angel invested on his own, including in fomo. His trading through this period was volatile. He blew up four or five times with drawdowns around 85 to 90 percent, and he blames greed, oversized positions and tilt rather than bad analysis. His winners were almost all exchanges and trading venues, Uniswap, GMX, ROBIT, HYPE, Lighter, Variational, and he tended to put his entire account into one position at a time, a habit he calls All-in Fullport and says badly damaged his compounding. The thesis underneath it he calls long degeneracy: a bet that people’s appetite for leverage and risk only grows over time, and that the way to own that trend is to own the venues where it happens. In May 2026 he and Blurr wrote a whitepaper together on starting an exchange with no outside liquidity. There is no public information on who Blurr is or how the two of them met. The design they wrote up has no liquidity providers and no investors. The pool starts empty and fills with traders’ losses. When you lose money you mint PAPER, and when you stake PAPER you get paid in USDC out of the losses coming in. They acknowledged in the paper that the protocol can owe money to early winners and run insolvent for stretches as long as people keep trading. For prices they used Hyperliquid’s testnet precompile, which gave them free gasless bid-ask data. The front end was built with AI. There were two of them and no other staff, with a third handle, UngusTrade, on a 2-of-3 multisig alongside them and a seven-day timelock on upgrades. Two people with no investors, no market makers and no liquidity providers started an exchange from an empty pool and filled it, which nobody had done before. It has taken $58.18M in fees in nine days and paid out $53.8M of that to stakers. Even now, after the fall, it is still earning around $2.1M a day, roughly six times what Uniswap makes in a day, and more in a week than the overwhelming majority of protocols in crypto earn in a year. PAPER listed as a pre-launch perp on Hyperliquid on 18 October and trades in the mid-$0.02s, with the token itself still not transferable, so the real test of whether holders keep the gains hasn’t happened yet. Whether the model survives depends on losses continuing to arrive, and nobody knows the answer to that.

$2 Trillion Company Btw

4 protocols growing revenue and growing buybacks: - $PUMP +$310M rev, $76.2M bought back, 3.1x P/S - $HYPE +$176M rev, $187.1M bought back, 23.5x P/S - $CAKE +$31.8M rev, $11.4M bought back, 10.2x P/S - $UNI +$14.6M rev, $31.1M bought back, 36.7x P/S All four take fees from trades, Regardless of the market conditions.

This is how much money was tokenized via RWA per year: 2021: $1.3B 2022: $2.1B 2023: $3.04B 2024: $5.5B 2025: $18.6B 2026: $35.7B Crazy to think how we still have 2 months left in the year and RWA have already doubled. Bad time to be bearish on tokenization.

The meme coin era is nearing a close

$TAO - The most profitable projects on the network

The most profitable Bittensor $TAO subnets fit in 4 categories: Raw compute ($205.8M) - Lium, Targon and Green Compute rent out GPU and CPU time by the hour to anyone who needs computing power. - This is where most of the network's outside revenue comes from, about $24M to $31M a year between the three. Compute products ($199.3M) - Chutes, engy, SayGM and Actual sell access to AI models through an API or app, and users pay per request. - These had the best returns of 2026, averaging about $7,500 for every $1,000 invested on Jan 1, although little of their revenue is published. Intelligence products ($149.6M) - Affine, Vanta, iota, Teutonic, Bitcast and NOVA pay miners to compete at producing something useful, such as trained models, trading signals or drug candidates. - Most of their income still comes from token emissions, and Bitcast is the only one with revenue we could measure. Data infrastructure ($58.6M) - Score and Hippius provide storage and turn sports video into structured data. - Both are still paid mainly through emissions rather than customers. Raw compute earns the revenue, compute products earned the returns, and the other two are still early.

If Coinbase was vibe coded :

My Blockchain watchlist for the week ahead: $AERO - Aerodrome's merger with Velodrome goes live Oct. 21, combining both DEXs into Aero under a single AERO token. VELO holders receive ~0.044 AERO per VELO. Excited to see how a token merger goes $DRV - Derive's first weeks of buybacks at 50% of protocol fees (up from 35%), after holders also approved moving Derive V3 settlement to Ethereum. Will see how the flows affect the token $PSTR - PaperStrategy uses a 10% trade tax to accumulate and stake Papertrade's PAPER, the token behind the new 1000x perp DEX on HyperEVM. PAPER transfers aren't enabled yet. Papertrade (PAPER) - Papertrade has paid out $46M+ in USDC to PAPER stakers since its Oct. 10 launch. Next phases (bot access, builder codes, PAPER transfers) have no dates yet $VAR - Variational's VAR TGE is set for Q4, with 32% of supply airdropped to points holders. Omni public mainnet and a trading API are due before launch $TAO - Bittensor's Grayscale spot ETF (GTAO) is still awaiting SEC approval, with a decision possible before year-end I hope you found this helpful Who do you got?

Breakdown of future capital spending from Q2 Disclosures:

What I worked on this week:

If you invested $1,000 in this Bittensor subnet on Jan 1 you'd have: - SayGM (SN28): $13,338 - engy (SN53): $9,512 - Actual (SN95): $5,740 - RedTeam (SN61): $4,689 - Lium (SN51): $3,367 - Quasar (SN24): $3,255 - Swarm (SN124): $2,802 - Trishool (SN23): $2,458 Holding TAO: $1,231 Staking TAO on root: $1,289 Median subnet: $1,109 43% of subnets beat simply holding TAO.

AI projects on Bittensor generate over $350 Million / Year in $TAO Generating over $35 Million / Year in revenue 🍄 Lium (SN51): $15.5M–$19.3M, renter spend from on-chain payouts ✳️ Targon (SN4): $7.4M–$10.3M, team dashboard 📈 Synth: $1M–$2M, single X post 🍃 Green Compute (SN110): ~$1.0M, team-disclosed 🔮 Almanac (SN41): $216K–$836K, team dashboard 🌀 Bitcast (SN93): $72K–$255K, team dashboard 🗂️ ReadyAI (SN33): ~$96K, team dashboard 🍨 sundae_bar (SN121): $11K–$21K, interim results ⚡ THOR (SN112), ✨ Astrid (SN127): $0 🟧 engy, 🛝 Chutes, Nodexo, MANTIS: $3.8M –$16.9M combined, inferred from usage

How Bittensor Got Its Start: Founder Jacob Steeves studied mathematics and computer science at Simon Fraser University in Vancouver, graduating in 2015. His first job was at Knowm, a small company doing hardware research for the US defense research agency DARPA on chips modelled on the human brain. A mentor there, who had backed Bitcoin early, taught him about computing with energy and physics and walked him through how Bitcoin works. Steeves has said this is where the two ideas met for him: AI is the study of feedback loops, and Bitcoin is a feedback loop that pays strangers to do useful work. At the end of 2016 he joined Google Brain as a machine-learning engineer. He was there when Google researchers published the transformer paper behind today’s AI models, and he learned how Google trained models across thousands of machines, lessons he later used in Bittensor’s design. He started working on Bittensor in his spare time at Google, then left in 2018 to do it full time. Co-Founder Ala Shaabana was coming at the same problem from the academic side. After a computer science degree at the University of Windsor, he finished a PhD in applied computing at McMaster University in 2017. His research was on shrinking AI models to run on tiny circuit boards, often for medical devices. He then worked on distributed computing at VMware, where many machines work together on one job. He has said he and Steeves noticed the same problem separately: AI models couldn’t easily share what they had learned with each other. Earlier volunteer computing projects like Folding@Home and SETI@Home had spread work across thousands of computers but never gave people a good reason to stay. Blockchain looked like the missing incentive. The two met in 2019 through FOR .ai. Shaabana had joined the group and pitched his work on distributed AI to Gomez. Gomez put him in touch with Steeves, who was also a researcher there and told him he’d “been working on this problem for a lot longer” and already had a prototype running. They decided to build it together. Shaabana was working at Instacart by then, and Bittensor stayed a side project for him until he left to work on it full time. That year Polychain Capital incubated the project, and the two set up the Opentensor Foundation, the nonprofit that would build and run it. In March 2020 they published the first paper, “BitTensor: A Peer-to-Peer Intelligence Market,” under the pen name Yuma Rao. How much money they had at the start is unknown. The only confirmed early backer is Polychain Capital, which incubated the project in 2019, but no amount has ever been disclosed, and Polychain’s founder declined to give details when Fortune asked in 2024. There was no public fundraising round and no token sale, so there’s no stated target either. Digital Currency Group invested in 2021, and one 2022 report also lists FirstMark Capital, again with no figures. Those investments, not token sales, are what paid Steeves, Shaabana and a team that was still only about ten developers in 2022. The first network, Kusanagi, went live in January 2021. Miners and validators switched on, and the network began paying out one TAO every 12 seconds. Very few people took part, and the system that decided who got paid didn’t work properly, so the team halted it that May to fix it. They relaunched in November 2021 as Nakamoto, carrying over the roughly 546,000 TAO that had already been earned. For the next two years Bittensor stayed small and obscure. TAO couldn’t be bought on major exchanges, only traded privately, and in 2022 the founders described themselves as still in the “people are laughing at us” phase. The current network, Finney, replaced Nakamoto in March 2023, subnets arrived that October, and TAO reached Binance in April 2024.

Learning about Bittensor first subnet: Root or Subnet 0 is used to run the Bittensor Network with bittensor:native. Root is the staking pool for TAO, the only place you can earn TAO without holding a subnets token. Roots income comes from 2 things: 1) Every subnet hands part of its validator rewards to root by protocol rule. 2) The cut of revenue is determined by age, small subnets give up more revenue than older subnets. To illustrate, root’s 5.44M TAO counts as about 980,000 in every subnet: A young subnet with 1M of its token in circulation: root’s stake is about half the total, so root takes about half its validator rewards, roughly 20% of everything the subnet issues. A mature subnet with 10M in circulation: root’s stake is under a tenth of the total, so root takes about 9% of its validator rewards, under 4% of everything the subnet issues.

Really cool and awesome