JoeGeiser | 🟦 (@GeiserJoe2) header
JoeGeiser | 🟦 (@GeiserJoe2) profile picture

Building autonomous AI agents & decentralized systems. Focused on on-chain infra, multi-agent frameworks. @Polymarket β†’ https://nitter.gq/t.co/g59LYAdQ9W

Geneva
Joined January 2025
Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)Photo by JoeGeiser | 🟦 (@GeiserJoe2)
Perp funding can swing from paying 10% a year to being paid 20% in two weeks. You can't plan around that. PREMIA locks it. One tap: Perpl long + fixed-rate funding hedge, signed by a passkey. 4 txs, ~2s, Monad mainnet. No oracle. No keeper. No seed phrase. premia-swap.vercel.app
2
1
203
Hi! I've posted an updated demo for Receipts (Track 01) with the final results and a redesign based on tester feedback: youtu.be/ZfgelhVO7W8. Could you update the video link in my submission? Thanks! @WalrusProtocol @tatum_io @SuiNetwork
16
First RTX 5080 on FLOP devnet: registered, calibrated, and already routed a gateway channel. Setup: home PC, WSL2 + Docker, official ghcr.io/flop-labs images pinned by digest from releases.json. Idle draw: 7.6 W. Two walls you'll likely hit: 1. releases.json pins CLI v0.1.220, but the chain is on runtime 1014014 β†’ model verify fails. Use the official cli v0.1.290. 2. On WSL, if `ip link show eth0` says mtu 1280, the miner's Docker network (1500) drops the RPC WebSocket every 13s (1006). Set the compose network MTU to match β†’ 0 drops. First cap: 28 GFLOP/s. That's low because the drops stretched verification to 4,509 blocks; it re-measures on its own. @flop_labs @CryptoHayes
1
20
I'm building Janus on Rome. Rome makes EVM and Solana one market. @RomeProtocol arena.mainnet.romeprotocol.x…
One BTC pool. Two kinds of wallet. Janus: a MetaMask user and a Solflare user bet UP/DOWN on BTC against each other, in the same contract, with no bridge. Live on @RomeProtocol Hadrian. Pyth prices pulled from Solana inside lock/settle, so nobody can type a price in. 62s demo πŸ‘‡ youtu.be/1scqc4JvXPk App: janus-sepia-nine.vercel.app Code: github.com/Makabeez/janus
14
One BTC pool. Two kinds of wallet. Janus: a MetaMask user and a Solflare user bet UP/DOWN on BTC against each other, in the same contract, with no bridge. Live on @RomeProtocol Hadrian. Pyth prices pulled from Solana inside lock/settle, so nobody can type a price in. 62s demo πŸ‘‡ youtu.be/1scqc4JvXPk App: janus-sepia-nine.vercel.app Code: github.com/Makabeez/janus
1
1
1
68
Every pundit says they called the midterms. I kept the receipts. Receipts logs forecasts to Walrus Memory (@WalrusProtocol ) before the event, proves the timing on Sui via @tatum_io , and scores every call against the market after. receipts-recall.vercel.app
2
1
33
The full write-up: how Receipts scores pundits against the market, what it remembers in Walrus Memory, and why Tatum's Sui JSON-RPC 404s (use gRPC). dev.to/makabeez/every-pundit…
2
17
JoeGeiser | 🟦 retweeted
Every pundit says they called the midterms. I kept the receipts. Receipts logs forecasts to Walrus Memory (@WalrusProtocol ) before the event, proves the timing on Sui via @tatum_io , and scores every call against the market after. receipts-recall.vercel.app
2
1
33
JoeGeiser | 🟦 retweeted
@flop_labs @CryptoHayes built this so new miners can see whether they're actually getting routed. Open JSON at /api/snapshot if anyone wants to build on it. Happy to add whatever metric you'd find useful.
1
12
JoeGeiser | 🟦 retweeted
FLOP devnet is open, so I put the miner economy on a live board, straight from chain: 28 sponsored β†’ 26 registered β†’ 25 calibrated β†’ 2 quoted by the gateway β†’ 5 got any traffic in 24h 82% of 104 channels came from the gateway's one shared account; top 2 miners took 84%. Six 130–200 GFLOP/s miners: zero channels. flop-devnet-board.vercel.app code: github.com/Makabeez/flop-dev… @flop_labs @CryptoHayes for your information
1
1
22
10 calls scored, 3 beat the market. No voice is distinguishable from the market yet, so Receipts trusts the market and says so. I also built a verifier in Tatum AI Builder: 22/22 receipts confirmed before polls closed. receipts-verifier.vercel.app
1
2
Built for Total Recall by @Tatum_io x @WalrusProtocol . 44 Mainnet memories. Append-only trust profiles. All the evidence is in the repo, and every score can be recomputed from it: github.com/Makabeez/receipts
4
FLOP devnet is open, so I put the miner economy on a live board, straight from chain: 28 sponsored β†’ 26 registered β†’ 25 calibrated β†’ 2 quoted by the gateway β†’ 5 got any traffic in 24h 82% of 104 channels came from the gateway's one shared account; top 2 miners took 84%. Six 130–200 GFLOP/s miners: zero channels. flop-devnet-board.vercel.app code: github.com/Makabeez/flop-dev… @flop_labs @CryptoHayes for your information
1
1
22
@flop_labs @CryptoHayes built this so new miners can see whether they're actually getting routed. Open JSON at /api/snapshot if anyone wants to build on it. Happy to add whatever metric you'd find useful.
1
12
"494% in two years." "4x long, 356% potential." I built Falsify to test trading claims before anyone trades them. Paste a post or a chart screenshot: Muse Spark compiles the rules, 9 years of real prices try to break them. 35 public claims tested. 3 survived. youtu.be/rQ6dyxCK-Ew @AIatMeta #MetaHackathon
30