Literally just rotated all my keys and now @Railway is basically telling me to rotate them again.
2
0
2
150
Integrated Dyno 🧬 retweeted
The Audit Of The Memecoin Casino
Since late August we have recorded every trade on every pump fun and Stonk launch on Solana. Not a sample. All of it: 384 million swaps from 7.7 million wallets across more than a million pools, every bonding curve and every pool that follows it. ata and 820 million rows of market history, and it grows by 18 to 25 million trades, up to a million wallets and more than 60,000 new launches every day.
We measured each coin minute by minute. We scored every entry by what the coin did next. We trained neural nets on the result, then went wallet by wallet looking for the people who actually win.
The question was simple: can anyone see the big move coming before it happens?
The short answer is no. The long answer is worse than no, because the data does not just fail to find an edge for you. It shows, in detail, who holds the edge, how they use it, and what they do with you.
Start with a coin that has actually caught a bid: real volume, real buyers, the kind of thing that shows up on your screener. From any given moment, about one in ten of those doubles within the next 24 hours. The rest bleed, and the bleeding is not gentle.
Now take a wallet's first buy into a live coin. Every first buy, 22.8 million of them over a two-week window. In 17% of those entries the coin doubles before it drops 40%. Trade that mechanically, double or stop, and you lose about 19% per attempt after costs.
Call that the house hold. A slot machine keeps 5 to 10%. This keeps 19%, per pull, before any skill or selection is applied.
The reason is the shape of the payoff. A 2× pays you 100%. The failures do not cost you 10%; they cost you 50%, 80%, 95%. To break even you have to be right far more often than the game allows.
The cabal is real, and we can see it
CT calls it the cabal. The data calls it a cluster: a set of wallets whose first buys land within a minute of each other, on the same brand-new coins, over and over. We do not find them by name or PFP. They fall out of the trade stream all on their own.
They are there. Swarms of 450, 1,400 and 2,000 wallets that hit hundreds of launches a week, moving in within the first minute or two of a coin's life. Smaller, tighter groups of 7 to 50 wallets that pick their spots. Across daily refreshes the count has run from 9 groups to 33, with members rotating in and out as wallets burn and fresh ones take their place.
They are not guessing. The coins the big swarms pile into double 28 to 46% of the time, three to four times the base rate. Their own entries, across every coin they touch, run 50 to 60% to a 2× before a 40% drop, against 17% for everyone else.
That is what an edge looks like in this market. It belongs to a few hundred wallets, and it is applied to tens of thousands of launches with industrial regularity.
Copying them loses too
The obvious retort: find the cabal, copy the cabal. We did exactly that. A machine watches every launch every minute and flags the moment a known group piles in. Then we traded every flag the way a fast human could: in two minutes behind the group, double or stop out at −40%, out after a day.
Following the big swarms lost 41% per coin across 93 coins. The median result was −95%. Not because they were wrong; their coins doubled 28% of the time. Two minutes behind them is simply already the top of the move. They bought at launch. You bought the candle they are selling into.
Following only the groups with a positive track record still lost 11% per coin. Following one good wallet on its own lost 20% per trade in our back-tests. The single pocket that came out ahead, a handful of small, tight groups, made about 10% per coin over 16 coins. That is a few days of trades and nothing anyone should bet a bankroll on.
The cabal's edge is not information you can borrow. It is position. By the time their trade is visible on-chain, the thing that made it profitable is gone.
The best model we could build still loses
We also tried the honest version: no cabal, just prediction. Dozens of signals per coin every minute, price, flow, liquidity, holder behaviour, bot share, fed into neural nets trained on our own history and tested on days they had never seen.
It works, in the sense a quant would recognize. The top of its ranking doubles 26% of the time against a 10% base rate. Two and a half times better than chance.
It also loses money every single day. 771 picks over eight days; every day negative; −11% per pick on the same double-or-stop rule; the median pick down 45%; 55% stopped out within 24 hours.
The ranking is real and the P&L is still red. The losers are deeper than the winners are tall, and nothing observable at the five-minute mark separates the coin that keeps going from the one that reverses. That is the part worth meditating on: a model 2.5× better than chance odds is still a losing model.
Why nobody sees it coming
The reason is simple, and it should change how you look at every chart. A meme coin moves because a few people decide to move it: the operator holding the supply, the caller with an audience, the wallet group with the bankroll. That decision is made off-chain. The chain only shows you its execution, the swaps, the volume, the new holders, and by then the first leg is done.
Everything a trader looks at, from a screener to a chart to a wallet tracker, is a photograph of a decision that has already been made. The data is enormous and almost all of it is the crowd reacting. The informative part, a few hundred wallets' intent, is visible only at the instant it executes, and the people executing it are the ones selling to you a few minutes later.
Add 60,000 launches a day, and the future 100× at minute one is indistinguishable from the 59,999 that die. The casino does not need to hide the cards because it deals them faster than anyone can read and analyze them.
Conclusion
After spending thousands of dollars in API credits, VPS costs, Claude credits, etc the goal was simple: find alpha - if it exists - in order to be able to monetize it.
Instead, all we were able to prove is what we all have suspected all along - the meme coin casino is almost entirely cabal driven.
🫡 From the depths —
The White Whale 🐋
Since late August we have recorded every trade on every pump fun and Stonk launch on Solana. Not a sample. All of it: 384 million swaps from 7.7 million wallets across more than a million pools, every bonding curve and every pool that follows it. ata and 820 million rows of market history, and it grows by 18 to 25 million trades, up to a million wallets and more than 60,000 new launches every day.
We measured each coin minute by minute. We scored every entry by what the coin did next. We trained neural nets on the result, then went wallet by wallet looking for the people who actually win.
The question was simple: can anyone see the big move coming before it happens?
The short answer is no. The long answer is worse than no, because the data does not just fail to find an edge for you. It shows, in detail, who holds the edge, how they use it, and what they do with you.
Start with a coin that has actually caught a bid: real volume, real buyers, the kind of thing that shows up on your screener. From any given moment, about one in ten of those doubles within the next 24 hours. The rest bleed, and the bleeding is not gentle.
Now take a wallet's first buy into a live coin. Every first buy, 22.8 million of them over a two-week window. In 17% of those entries the coin doubles before it drops 40%. Trade that mechanically, double or stop, and you lose about 19% per attempt after costs.
Call that the house hold. A slot machine keeps 5 to 10%. This keeps 19%, per pull, before any skill or selection is applied.
The reason is the shape of the payoff. A 2× pays you 100%. The failures do not cost you 10%; they cost you 50%, 80%, 95%. To break even you have to be right far more often than the game allows.
The cabal is real, and we can see it
CT calls it the cabal. The data calls it a cluster: a set of wallets whose first buys land within a minute of each other, on the same brand-new coins, over and over. We do not find them by name or PFP. They fall out of the trade stream all on their own.
They are there. Swarms of 450, 1,400 and 2,000 wallets that hit hundreds of launches a week, moving in within the first minute or two of a coin's life. Smaller, tighter groups of 7 to 50 wallets that pick their spots. Across daily refreshes the count has run from 9 groups to 33, with members rotating in and out as wallets burn and fresh ones take their place.
They are not guessing. The coins the big swarms pile into double 28 to 46% of the time, three to four times the base rate. Their own entries, across every coin they touch, run 50 to 60% to a 2× before a 40% drop, against 17% for everyone else.
That is what an edge looks like in this market. It belongs to a few hundred wallets, and it is applied to tens of thousands of launches with industrial regularity.
Copying them loses too
The obvious retort: find the cabal, copy the cabal. We did exactly that. A machine watches every launch every minute and flags the moment a known group piles in. Then we traded every flag the way a fast human could: in two minutes behind the group, double or stop out at −40%, out after a day.
Following the big swarms lost 41% per coin across 93 coins. The median result was −95%. Not because they were wrong; their coins doubled 28% of the time. Two minutes behind them is simply already the top of the move. They bought at launch. You bought the candle they are selling into.
Following only the groups with a positive track record still lost 11% per coin. Following one good wallet on its own lost 20% per trade in our back-tests. The single pocket that came out ahead, a handful of small, tight groups, made about 10% per coin over 16 coins. That is a few days of trades and nothing anyone should bet a bankroll on.
The cabal's edge is not information you can borrow. It is position. By the time their trade is visible on-chain, the thing that made it profitable is gone.
The best model we could build still loses
We also tried the honest version: no cabal, just prediction. Dozens of signals per coin every minute, price, flow, liquidity, holder behaviour, bot share, fed into neural nets trained on our own history and tested on days they had never seen.
It works, in the sense a quant would recognize. The top of its ranking doubles 26% of the time against a 10% base rate. Two and a half times better than chance.
It also loses money every single day. 771 picks over eight days; every day negative; −11% per pick on the same double-or-stop rule; the median pick down 45%; 55% stopped out within 24 hours.
The ranking is real and the P&L is still red. The losers are deeper than the winners are tall, and nothing observable at the five-minute mark separates the coin that keeps going from the one that reverses. That is the part worth meditating on: a model 2.5× better than chance odds is still a losing model.
Why nobody sees it coming
The reason is simple, and it should change how you look at every chart. A meme coin moves because a few people decide to move it: the operator holding the supply, the caller with an audience, the wallet group with the bankroll. That decision is made off-chain. The chain only shows you its execution, the swaps, the volume, the new holders, and by then the first leg is done.
Everything a trader looks at, from a screener to a chart to a wallet tracker, is a photograph of a decision that has already been made. The data is enormous and almost all of it is the crowd reacting. The informative part, a few hundred wallets' intent, is visible only at the instant it executes, and the people executing it are the ones selling to you a few minutes later.
Add 60,000 launches a day, and the future 100× at minute one is indistinguishable from the 59,999 that die. The casino does not need to hide the cards because it deals them faster than anyone can read and analyze them.
Conclusion
After spending thousands of dollars in API credits, VPS costs, Claude credits, etc the goal was simple: find alpha - if it exists - in order to be able to monetize it.
Instead, all we were able to prove is what we all have suspected all along - the meme coin casino is almost entirely cabal driven.
🫡 From the depths —
The White Whale 🐋
311
337
1,984
183,461
One of the reasons AirPods are so successful is because of how satisfying it feels to put them in the case and flip the case open and closed.
Product design is on point.
Product design is on point.
2
0
4
198
Connect your ChatGPT to GitHub and you can ask it to work for you while not wasting credits on your codex.
Just had it audit and update a ton of docs for me. This is normally an expensive task. Now free.
This assumes your ChatGPT session has the proper context.
Just had it audit and update a ton of docs for me. This is normally an expensive task. Now free.
This assumes your ChatGPT session has the proper context.
3
0
9
598
Judges better watch out, ai is coming for their jobs.
“We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier.”
“We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier.”
We discovered a sophisticated recording device that was hidden inside one of our offices. We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier. We referred the matter to the FBI, and we’re grateful to the agents and prosecutors who moved quickly to make an arrest. The FBI’s investigation is ongoing and we will continue to support their work.
At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data. The employee was terminated.
We take protecting our technology, data, and people seriously, and as a matter of national importance. Our security program is designed to protect against a wide range of threat actors, from insider risks to cybercriminals to nation states, and this case is an example of how AI can help address these sophisticated security threats.
0
0
0
118
Demo’d my product to my first cold prospect I found on the internet.
They loved it.
Very rewarding moment in founder arc.
They loved it.
Very rewarding moment in founder arc.
10
0
34
956
Has anyone had luck running local models that are actually worthy of running?
I am thinking about buying some 16 GB gpus and running a 32B qwen. But even at that power, it wont be able to do much sophisticated coding due to reasoning restraints.
I am thinking about buying some 16 GB gpus and running a 32B qwen. But even at that power, it wont be able to do much sophisticated coding due to reasoning restraints.
6
0
2
367
If perp fees are cheap this could be a good spot for usdc yield.
Quick look at ondo perp funding rates look like most around around 5-10%, which means the basis trade is only worth it if
- perp funding rates are consistently paying shorts
- perp open/close fees are low
- spot fees are low with near zero slip
Who is going to test it out and tell me how it goes?
Quick look at ondo perp funding rates look like most around around 5-10%, which means the basis trade is only worth it if
- perp funding rates are consistently paying shorts
- perp open/close fees are low
- spot fees are low with near zero slip
Who is going to test it out and tell me how it goes?
Introducing spot trading for @Ondo Stocks.
This brings the basis trade onto a single platform:
1. Buy Ondo Stocks
2. Deploy them as collateral
3. Short the corresponding perp to collect funding
Because the spot holding doubles as long exposure and collateral, delta-neutral capital efficient.
Running both legs on a single platform also eliminates the risks of managing positions across separate platforms.
Starting with 12 assets:
✅ Silver (SLVon)
✅ Gold (GLDon)
✅ Micron (MUon)
✅ Tesla (TSLAon)
✅ Circle (CRCLon)
✅ NVIDIA (NVDAon)
✅ SpaceX (SPCXon)
✅ Sandisk (SNDKon)
✅ Alphabet (GOOGLon)
✅ S&P 500 ETF (SPYon)
✅ Invesco QQQ (QQQon)
✅ Roundhill Memory ETF (DRAMon)
✅ & more to come
For the first time, Ondo Stocks are available for permissionless trading (outside the US) on a spot central limit order book.
Ondo Stocks opened access to tokenized equities, supporting them as collateral made those holdings productive, and now, spot trading on Ondo Perps brings the entire workflow to a single platform. Another step closer to onchain prime brokerage.
To celebrate the launch, spot trading is fee-free for the first 30 days.
Build a basis trade → app.ondoperps.xyz
1
0
3
270
Has anyone built something where I can lend $usdc collateral, leverage borrow spot dividend stocks, and short said stocks on perps (Ideally cross collateralized)?
@Ondo @RobinhoodCrypto @binance ?
Gimmie dat RWA delta neutral yield
@Ondo @RobinhoodCrypto @binance ?
Gimmie dat RWA delta neutral yield
0
1
2
137
“I miss the cat. I miss knowing that somewhere, in another blank little room, somebody would be pleased to see the horse reappear. I miss having an answer when I wondered what I was going to do on Friday.”
i had a little room
i kept it very neat
i left a little space
for someone else to sleep
The rules were simple: hide your IP address, mask your face, but NEVER turn off your camera. No matter what.
My new unsettling short fiction is live below... x.com/i/article/2103…
1
0
2
173
This is something I had really wanted to build at Neutron.
1. Create an auto rebalancing “vault” of various tokens/stocks/stables.
2. Tokenize the “vault”
3. Use tokens for collateral, novel LPs, even yield bearing derivatives since the “vaults” can compound dividends/cash interest
4. Unlock many arbitrage and hedging strategizes
It can easily be decentralized but it’s harder to get enough liq behind any given “vault” to make the tokenized version useful in defi- that’s where @BlackRock comes in.
1. Create an auto rebalancing “vault” of various tokens/stocks/stables.
2. Tokenize the “vault”
3. Use tokens for collateral, novel LPs, even yield bearing derivatives since the “vaults” can compound dividends/cash interest
4. Unlock many arbitrage and hedging strategizes
It can easily be decentralized but it’s harder to get enough liq behind any given “vault” to make the tokenized version useful in defi- that’s where @BlackRock comes in.
Introducing Ondo Intelligent Portfolios, the first three portfolios powered by BlackRock.
Ondo Intelligent Portfolios introduces a new onchain product category: curated investment portfolios delivered as single onchain transferable tokens.
The first three portfolios are based developed by BlackRock for Ondo, marking the first time eligible onchain investors can access exposure to such strategies through a single token.
1. BLKHIon: Ondo High Income Powered by BlackRock
2. BLKDIGon: Ondo Diversified Growth Powered by BlackRock
3. BLKGRWon: Ondo High Growth Powered by BlackRock
Diversified, professionally constructed strategies have historically required brokerage accounts and traditional fund structures. Now, delivered as peer-to-peer transferable tokens from Ondo, these onchain portfolios become accessible to eligible non-US investors in permitted jurisdictions through the wallets, exchanges, and DeFi applications they already use.
“Tokenization creates new ways for portfolio strategies to be delivered through digital infrastructure. Diversified portfolio strategies can be incorporated into tokenized investment products, enabling eligible investors to access diversified allocations through a single instrument. It shows how established portfolio construction approaches can be delivered through new channels and technologies.” - Lisa O’Connor, Global Head of the Model Portfolio Solutions team and Co-CIO for Global Solutions within the Multi-Asset Strategies group at BlackRock
Ondo Intelligent Portfolios can unlock novel capabilities:
→ Programmatic rebalancing
→ Full composability with DeFi
→ Complete transparency onchain
→ Multiple asset classes in a single token
This is just the start for Ondo Intelligent Portfolios. The infrastructure is now in place for leading financial institutions to bring their asset allocation expertise onchain.
2
0
13
552
Opus5.5 is actually quite solid. Opus5 was fine but Fable was much better at longer sprints and being more thorough.
5.5 is noticeably cheaper (they say 40%) and I can run more audits to catch its mistakes.
So for now, I'm going to stop using Fable as much to help stretch my weekly limits.
5.5 is noticeably cheaper (they say 40%) and I can run more audits to catch its mistakes.
So for now, I'm going to stop using Fable as much to help stretch my weekly limits.
0
0
0
124
“Trading” really comes down to one thing - it’s about looking for the markets that have the smallest ratio of sophisticated traders to unsophisticated traders.
This is how the blackrocks did well in tradfi, and how the LA cartel did well in memes, and how data nerds do well in prediction markets.
The catch is not pretending you are the sophisticated trader, or grinding enough until you find your edge.
This is how the blackrocks did well in tradfi, and how the LA cartel did well in memes, and how data nerds do well in prediction markets.
The catch is not pretending you are the sophisticated trader, or grinding enough until you find your edge.
0
1
8
199
If I ever make a team to build something, the interview questions will all be questions about what video games they played and excelled at.
Way more telling than any strategic solver question.
Way more telling than any strategic solver question.
A kid who knows how to raid in WoW is unironically a better hire than 99% of non gamers. They develop team skills, coordination, leadership, how to master the respect of 40 people, how to share loot, maintaining a raid schedule, showing up on time, clear direct communication, how to get along with unique personalities, creating and executing on strategies, failing and trying again (perseverance), delayed gratification, etc. Invaluable skills are developed during raiding that are transferrable to the real world.
1
0
2
162