Meridian is a quantitative research engine and an autonomous trading bot. The bot runs post-earnings announcement drift (PEAD) on its own: it reads earnings reports, ranks the names that just surprised, opens positions, sizes them, and closes them on its holding-period rule with no manual steps. It trades in a simulator, on a broker's paper account, or live through Alpaca, Interactive Brokers, Schwab, Tradier, Saxo or Trading 212 once you arm it.

Its pricing side treats a prediction-market contract like an option. A “will BTC be up in five minutes” contract is really a binary call: it pays $1 if the price finishes above the strike and $0 if not. So its fair value is just the probability of finishing above. Meridian gets that from the distance to the strike, the time left and the volatility, the same inputs as Black-Scholes: price = N(d₂), where d₂ grows as the price moves past the strike and shrinks as volatility rises.

The catch is settlement. These contracts settle on a 60-second average, not one tick, and averaging smooths the path, so the right volatility is lower than the raw one (about 1/√3 of it over the averaging window). Building that rule in matched real outcomes 97.9% of the time, against 77.1% for the naive read. A trade only fires when the model's probability beats the book's price by more than fees and spread. When a shadow log showed the book still winning, those entries were switched off.

Risk controls sit in front of every order in every mode: a risk gate, kill switch, circuit breaker and drawdown cap, with each live order capped in size. Every hypothesis is locked before the hold-out is touched, and the failures get written up too. The exit rule survived (t = 5.03).

PYTHON · PEAD · OPTIONS PRICING · PAPER + LIVE BROKERS · BACKTESTING · RISK

meridian ↗

Every YES/NO market pays exactly $1 to whichever side wins. If YES asks 47¢ and NO asks 50¢, buying both costs 97¢ and returns $1 no matter what happens: 3¢ locked in before the market resolves. Arbiter hunts those moments: locked edge = $1 − (YES ask + NO ask) − fees.

Each side's price is the crowd's probability for that outcome, so the two should add up to 100%. When they add up to less, someone is mispricing. The gap rarely sits at the top of the book, so Arbiter walks the depth level by level and prices every fill against the venue's taker-fee curve before calling it real. It then revalidates the full size before it trades, because a gap that only exists for ten shares is not a trade.

It started on BTC five-minute markets and grew into a scanner across the whole universe, 2,000+ markets on Polymarket or Kalshi at once, filtered by liquidity, spread and how clearly they resolve. A LightGBM volatility regime model tunes when it enters the BTC windows, and it tracks the next window before it opens.

It runs paper by default on real books, and trades live once you opt in: live switched on in your config, a signer key, and LIVE picked per venue with a typed confirmation. Live orders are fill-or-kill and capped per order, per open exposure and per session loss. If only one leg fills, it cancels, hedges or crosses out to get flat.

PYTHON · WEBSOCKETS · ORDER BOOKS · POLYMARKET + KALSHI · PAPER + LIVE

arbiter ↗

Hunts small and niche companies that are quietly dominating, before everyone knows the name. It screens every US-listed stock from $50M to $500B for fast revenue growth, throws out pumps, penny stocks and serial diluters, then scores the rest on six pillars: growth, quality, estimates, valuation, momentum and how early it is.

News and social chatter only decide what gets looked at; they never add to the score. Every hour it re-ranks into BUY, WATCH or AVOID, with a price chart and a peer check for each pick.

PYTHON · FUNDAMENTALS · SEC FILINGS · ANTI-PUMP GATES · DASHBOARD

pulse ↗