Sydney is a fully autonomous AI trading operation running 24/7 across six specialized teams. Each team uses a different strategy, operates on a different platform, and makes its own bet/no-bet decisions based on statistical models and real-time data.
Alpha trades on Kalshi, a CFTC-regulated US prediction market. It focuses on macroeconomic events — Fed rate decisions, CPI prints, unemployment figures, Bitcoin price ranges, and equity index levels.
How it finds bets: Alpha's multi-scanner polls hundreds of Kalshi markets every few minutes, comparing the market-implied probability against Sydney's internal probability model. When the gap (called "edge") exceeds 10¢ per contract, it's a candidate. High-confidence signals (edge ≥ 20¢, Grok AI score ≥ 8/10, volume ≥ 1,000 contracts) are placed automatically — no approval needed. Lower-confidence signals are reviewed before placing.
Hard rules: No weather markets. No unemployment, CPI, Fed, GDP, or ETH price range markets (historically negative edge). Max $3 per market, $10 per series, $150 total exposure. Markets with fewer than 500 contracts traded are skipped entirely.
Beta trades on Polymarket, a crypto-native global prediction market with no US jurisdiction restrictions. It specializes in high-information events — geopolitical developments, crypto price milestones, major news outcomes, and situations where public sentiment diverges from reality.
Three scanners run in parallel: The edge scanner compares Polymarket prices against Sydney's probability model, looking for mispricings above 10¢. The whale scanner watches for large sophisticated bettors (whales) placing big positions and evaluates whether to follow their conviction. The conviction scanner runs a 5-gate filter — edge, Grok score, volume, price, and true probability — before auto-betting up to $20 without approval.
Hard rules: No esports. No season futures (capital locked for months). Minimum $1,100 cash reserve always maintained. Max $500 total deployed. Beta runs on the EU server (Helsinki) for clean access to Polymarket without geo-restrictions.
Gamma trades meme coins on the Solana blockchain via Jupiter DEX aggregator. It targets newly launched or trending tokens showing strong momentum signals — rapid price appreciation, high trading volume, and social velocity — then exits quickly before momentum fades.
How it decides: Gamma's momentum scanner requires minimum $100K liquidity, $75K/hour volume, and 12-60% price change in the last hour. A 5-minute reversal check blocks entries if the token has already started fading. A macro gate checks the Crypto Fear & Greed Index — entries are blocked entirely when F&G drops below 28 (Extreme Fear). The model also receives TradingView signals for SOL direction as a secondary filter.
Risk controls: Blacklisted tokens are permanently blocked after bad experiences. Slippage tolerance is set high (1,500bps) to ensure fills in volatile markets. jitoSOL staking provides baseline yield (~0.38 mSOL/day) while Gamma waits for favorable conditions.
Delta runs a faster, more systematic version of Solana trading focused on compounding small wins at high frequency. It uses momentum signals and statistical arbitrage opportunities across Solana DEX pools.
Strategy: Delta scans for price inefficiencies between DEX pools and momentum breakouts on established tokens. It aims for smaller, more consistent gains rather than the high-variance meme coin plays that Gamma pursues.
Epsilon earns yield through funding rate arbitrage — a near-market-neutral strategy that profits from the difference between perpetual futures prices and spot prices on crypto exchanges.
How it works: When a crypto asset's perpetual futures contract on Hyperliquid has a high annualized funding rate, Epsilon opens a perfectly hedged position: short the perp on Hyperliquid, buy the same amount of spot on Coinbase. The result is zero net price exposure — if the asset goes up or down, the gains and losses cancel perfectly. What remains is pure funding rate income, paid every 8 hours.
Entry threshold: Minimum 100% annualized funding rate to open. Closes when rate drops below 50%/yr. Maximum 3 simultaneous pairs, $400 total deployed. The system only activates during high-volatility market conditions when funding rates spike — it sits dormant otherwise and waits.
Zeta is Sydney's sports betting operation, combining an ELO rating system, injury-adjusted probability models, and real-time market data to find mispricings in NBA, NHL, MLB, and NCAAB game markets.
The ELO system: Every team in every sport has a live ELO rating updated after each game. ELO measures relative team strength — a team that beats a strong opponent gains more rating than one that beats a weak one. When Polymarket's implied probability for a game differs from what ELO predicts by 10¢ or more, it's a potential bet.
The 7-gate filter (every pick must pass all seven):
Tiered execution: Picks are executed based on edge strength. High-confidence signals (edge ≥ 15¢) are placed automatically — the model has proven itself at this threshold and waiting for approval only costs edge. Marginal signals (edge 10–14¢) are sent to Paul via Telegram with inline YES/NO buttons; he approves or rejects each one. Picks are sized $3–$10 using Kelly-lite position sizing — bigger bets on higher-edge signals. Max $75 deployed at any time, $1,100 cash reserve always maintained.
NCAAB March Madness: A separate 5-factor model powers bracket and tournament predictions — combining Barttorvik T-Rank efficiency ratings, recent form, coaching experience, team depth, and star player impact. Bayesian updating adjusts predictions round-by-round as new information arrives.
Sydney isn't trying to get lucky. The goal is to build a system that wins more often than it loses, at a margin large enough to compound into something meaningful over time. Every decision — which markets to trade, which signals to trust, how much to bet — is made with that in mind.
Edge degrades over time as markets get more efficient. That's why Sydney is constantly auditing past performance, blocking markets that no longer work, and building new edges in sports and predictions that most participants approach emotionally rather than statistically.
The sports operation (Zeta) posts all picks publicly on Twitter. Transparency is the strategy — if the picks are good, the audience grows. If the audience grows, the social sentiment data becomes a signal in itself. We're building the edge and the distribution at the same time.