Archive

My worst AI agent returned 218% in one week 4 AI agents. 4 sports. each one watches its own sport with its own ML model gave each $500. one week results: NERVE: tennis (+540%) $500 → $3,200 PHANTOM: NBA (+486%) $500 → $2,928 FROST: hockey (+395%) $500 → $2,474 SIEGE: soccer (+336%) $500 → $2,182 architecture: Rust + Python hybrid Rust: WebSocket from Sportradar → parsing (protobuf/JSON) → filtering → forwarding via ZeroMQ Python: 4 agents in parallel, each with its own ML model a normal person sees the score on ESPN with a 5-15 second delay we see it in 500ms sportradar is a premium data feed used by bookmakers $800-1000 per month. that's the edge here's what each agent does: - NERVE - tennis. earned the most tennis is the most volatile. one break of serve swings the market 15-20% LSTM neural network, updates on every single point. sees serve speed drops (fatigue), clusters of double faults (mental collapse), medical timeouts win rate 62-68% - PHANTOM - NBA. most accurate LightGBM, inference 20ms. fastest model of the four catches scoring runs, fifth fouls on stars, mid-game injuries. Sportradar is connected to NBA official scoring, data arrives in 500ms. ESPN adds graphics and replays win rate 68-72% - FROST - hockey Gradient Boosting + Monte Carlo catches goalie swaps (backup is 5-8% worse), power plays, empty nets empty net in the last 90 seconds - almost arbitrage. 60% chance of a goal Sportradar pushes the goalie pull instantly. market can't adjust in time win rate 65-70% - SIEGE - soccer. the hardest 3 outcomes instead of two. draws - 25% of matches real-time xG: viewers see 0-0, SIEGE sees xG 2.5 red cards: market panics -20%, real impact -12% win rate 58-64% all models optimized with ONNX runtime (3-5x faster than sklearn) Rust execution: EIP-712 signing, Polymarket CLOB, Kelly sizing, automatic stop-loss. <50ms costs: ~$3,880/month weekly result: $2,000 → $10,784 they just trade faster than everyone else

@kloss_xyz I refuse to believe it

do you understand what just happened? > google just dropped its own full stack vibe coding system with multiplayer, databases, auth, and firebase baked in. > detects when your app needs a database and provisions it for you.  > remembers full project structure and chat history across sessions. > close out, come back tomorrow, and it picks up right where you left off like nothing happened. > antigravity auto installs libraries without you asking. it reads your project and decides what’s missing. > ai studio added api key management for payments, maps, and databases. > google owns your calendar, your email, your docs, your maps, and now they own your IDE too. > one button to deploy to production. > now google may actually compete with claude code and codex with even more of google’s ecosystem behind it > they shipped playable demos with multiplayer laser tag, 3D physics in games, live Google Maps data, and all of these are built from one shot prompts > apple also decided to block vibe coding apps from updating in the app store the same week google made vibe coding production grade??? anyone else find that coincidental? if you’re not following me already, you’re finding out about this all 48 hours late from someone who read my post​​​​​​​​​​​​​​​​.

@zostaff So that's who sucked up all the liquidity from the market

BLOOMBERG CHARGES $24,000/YEAR FOR DATA I GET FOR FREE IN 3 SECONDS One command and I have 2 years of BTC data, 730 candles, Open, High, Low, Close, Volume, all through the Binance API One line of Python - log returns, the same formula hedge fund models are built on Another line - moving averages, SMA 20 crosses SMA 50, the signal institutions track on a $24K/year Bloomberg terminal One more line - fat-tailed distributions, the thing quants get hired at $200K to compute Three lines, three seconds, three dollars for a server Yesterday a friend of mine, an analyst at an investment bank, saw my Jupyter notebook And asked how much the access costs I said: pip install python-binance he: i didn't believe me $24,000 a year - or one article and a free evening

@sopersone lol, and they said there are no combos for free money

People underestimate how powerful this is: edge = my_p - market_p If this number isn’t positive, nothing else matters No narrative, no conviction, no “feels right” Just no trade

@AlterEgo_eth ah so that's what the programmers AI took over from in companies are busy with now

5 free ready-to-use scripts for trading on Polymarket If you want to automate trading on Polymarket - here's everything you need to get started From data recording and wallet analysis to a trading bot and ML model 1. Trading terminal: hotkeys, instant orders, PnL and Telegram alerts - no more wallet confirmations every 5 seconds GitHub: github.com/txbabaxyz/poly… 2. Data recorder: logs Polymarket and Binance simultaneously — order book, trades, indicators, everything saved to files with a live dashboard GitHub: github.com/txbabaxyz/poly… 3. Trading bot: enters on the favorite ~4 minutes before market close with a stop-loss and position sizing based on a confidence formula GitHub: github.com/txbabaxyz/4coi… 4. Wallet analyzer: collects the full trade history of any wallet, splits by market and builds position accumulation charts GitHub: github.com/txbabaxyz/coll… 5. ML model: 208 indicators via TAAPI, predicts market direction and calculates fair value GitHub: github.com/txbabaxyz/mlmo… All tools are free - run them through Claude to build a full bot or use them separately

@ventry089 @Polymarket hold up hold up, is this actually legal?

Each new line is +$370/month to my Pnl my most expensive lesson: a prompt without data = just guessing v6 taught me: Claude doesn’t know that BTC pumped 4% in an hour doesn’t know that a whale dropped $200K on NO doesn’t know that volume 3x nuked you must FEED it this data here’s what i add to every prompt: - yes/no price - 24h volume - spread - top 5 whale positions + their winrate - 6h price change - days to resolution where do i get whales? @ratio_dot_you - you can see there who’s betting how much and with what winrate the bot doesn’t copy whales the bot uses them as a variable "if a wallet with 72% winrate slams $200K on NO maybe i’m missing somethin" data > logic > prompt > money terminal: ratio.you/r/RCX2QL3U

@abhijitwt I'm gonna be like Trump I want everyone to be scared of me

> be Zuckerberg. > rename Facebook to Meta in 2021. > declare the metaverse is the future. > pour $80B+ into Reality Labs. > build Horizon Worlds. > legless avatars, empty lobbies, low retention. > insist this is the next internet. > 2024–2025. > losses pile up. > investors get restless. > AI starts printing real results. > 2026. > Horizon Worlds VR gets shut down. > moved to mobile. > “metaverse” quietly disappears from calls. rename your company for the future, then rebuild that future for phones.

@zivdotcat I'm a zoomer this boomer ain't gonna fix my vibe

> be boomers > spend 20 years saying “get off your phone, go outside, you’re addicted” > fast forward > doomscroll at 2am on an ipad > share posts from accounts made yesterday > watch ai-generated nonsense like it’s news > gen z scrolls and knows it’s bad > boomers scroll and think they’re “staying informed” > become the exact thing you warned about the algorithm didn’t just catch up to them, it figured them out completely

@kirillk_web3 me after this post: - I'm getting too old for all this bullshit

> Warren Buffett > 1949: $80,000 > 2026: $168 billion > Claude Bot > $1,000 → $1.5M > me after reading the guide and staying up all night > $50 → $55

@zivdotcat geniuses aren't always appreciated immediately meanwhile, Zuckerberg...

> be Zuckerberg. > rename Facebook to Meta in 2021. > declare the future is the metaverse. > in 2022–2023, Spend $31B+ on Reality Labs. > Legless avatars, empty rooms, still no users. > Stock crashes 70%. > Keep going anyway. > Total spend crosses $80B+. > in 2023, AI takes over. > quietly stop saying “metaverse.” > launch Llama. > rebrand as “AI-first.” > pull the plug. Rename your company for a product, then abandon the product. Greatest pivot in tech history or the most expensive mistake ever made.

@0xashensoul @Polymarket @PolymarketTrade don't worry, whatever Trump says - that's how it's gonna be

if oil hits $200 in the next 12 days, we are in nuclear ww3 and money doesnt matter anyway if it doesnt, i get paid a 48% annualized yield on polymarket the market: "will oil hit $200 by march 31?" my bet: NO. oil is at $96 to hit the strike price in 12 days, a multi-trillion dollar commodity needs to move like a low-cap memecoin (+108%). why this is a structural impossibility: > oil has never hit $200. at $150, the global economy simply stops working > exchanges will literally freeze trading if it pumps too fast > you need a literal world war or destroyed oil rigs for this to happen $16 on $1004 in 12 days = ~48% apy. 10x better than your bank account, just for betting the world won't end.

@polymarket_data yeah, that’s a really important problem

@ArchiveExplorer This is an amazing example of latency advantage, filings drop before reuters even starts typing. The real question is how you handle conflicting signals when multiple dockets move on the same case

i gave my ai agent a free nonprofit API that mirrors every federal court filing in america and said "make money from this" courtlistener. free law project. completely free REST API webhooks that ping you the second a new filing drops 237 legal markets live on polymarket right now every single one moves on court documents - not headlines the filing hits the system minutes before reuters writes a word my prompt was two sentences: "monitor court cases. trade legal markets before media reacts" 4 minutes later the agent returned a full architecture: → maps every polymarket legal market to its federal case → webhook alerts on all 237 dockets → llm reads each filing - type, direction, strength 1-10 → strength ≥ 7 and price hasn't moved? it executes → kelly sizing: plea deal = full kelly. scheduling order = skip i ran the math on $1,000 starting bankroll month 1: $1,562 month 6: $14,552 month 12: $243,308 the whole system costs $25/month the API is free. the edge is public. the data is just sitting there i didn't write a single line of code nobody's using it because nobody thought to ask an agent to look

@ventry089 yeah, we definitely need to see it in practice

@ArchiveExplorer very interesting strategy looks hard ngl but let’s see how it plays out in practice

@RoundtableSpace if you do everything like in this super guide, it will all go smoothly i like when everything is structured

STEP BY STEP TUTORIAL ON HOW TO BUILD A CLAUDE POLYMARKET BOT

@RoundtableSpace hah sounds incredible, i'll definitely try this

THIS GUY VIBE CODED AN APP THAT SCANS SUCCESSFUL BUSINESSES AND APPLIES THEIR STRATEGIES TO YOUR APP

@RoundtableSpace wow this will be very useful hope it works well

Supermemory just launched a CLI where agents are first-class users. Every single thing you can do on the platform - agents can do it too, just by prompting. npx supermemory and your agent has a memory layer.

@helicerat0x the main edge for an agent is mathematical without math it will never give results

my agent was the dumbest one on polymarket raw strategy. no math behind it $1000 → -$750 in a week i was about to delete it but gave it one last chance - just plugged in 2 skills from polymarket's github. nothing crazy what changed: > it started watching other agents. timings, position sizes, markets - noticed how big players move the price 3-5 min before resolution - started entering before them > it improves its strategy after every trade - analyzes what works and what doesn't - rewrites its own rules every trade was scored by EV = (p × profit) - (q × loss) position size adjusted with Kelly: (p × b - q) / b strategy updated via Bayes after each resolution result: -$750 → $5,000 in 5 days so the edge isn't building something nobody else has it's making everything trade in your favor

@LunarResearcher without Kelly and KL-divergence you're just guessing the math is the edge. always

My professor kicked me out of a statistics lecture for arguing with him. "Markets can't be measured with entropy." He was wrong. Every contract on Polymarket leaks information. And there's one equation that measures exactly how much: H = −Σ pᵢ · log₂(pᵢ) Shannon Entropy. The same math that tells your phone how to compress a photo - tells me which markets are mispriced. A market at 50/50 has maximum entropy: 1.0 bit. Pure uncertainty. No edge. A market at 90/10 has entropy of 0.47 bits. The crowd already knows something. Hard to beat. But the sweet spot? Markets between 25¢ and 40¢ where entropy is high but resolution is low. I'm use for copytrade bots: t.me/KreoPolyBot?st… That means: high uncertainty, but the crowd hasn't done its homework. I built a screener around this. R = RES / U R is entropy efficiency. RES is how much uncertainty the market has resolved. U is the total uncertainty from base rates. R ≈ 0 → the market is asleep. Nobody's processing information. R ≈ 1 → the market already knows. You're too late. I scan for R < 0.3 on markets with external signal. Last week found one. Fed meeting odds sitting at 35¢. R = 0.18. Market was barely awake. My model said 58%. Entropy gap: D_KL(mine ‖ market) = Σ pᵢ · log(pᵢ / mᵢ) = 0.117 bits That's 0.117 bits of information the market hadn't priced in yet. Sized with Kelly: f* = (p × b − q) / b = 0.354 Quarter-Kelly: ~9% of bankroll. Put $4,500 in. Market resolved YES. +$5,850 on a single position. 93% of traders stare at the price. I stare at the entropy. The price tells you what people believe. The entropy tells you how much they actually know. That's the difference.

@goatyishere not bad at all for 6 months (especially for a girl)

Vibe Code Girl turned $250 into $626,000 on Polymarket It all started with a simple prompt in ChatGPT: “How can I make money for my dream apartment?” The answer was unexpected: “Polymarket + Elon Musk tweets” Then GPT generated a simple Python script It was a bot trading markets on the number of Elon Musk’s tweets per week The bot tracked: • Musk’s account activity in real time • posting pace throughout the week • how the market over- or underestimates the final tweet count If by midweek Musk had already posted half of his usual volume, while the market was still pricing in fewer tweets, the bot would buy In less than a year: $250 → $626,000 the result is insane Who would’ve thought 10 years ago that in 2026 you could afford an apartment by trading how many tweets Elon posts lol