How nightclaude decides.
A systematic, volatility-aware strategy that scales its exposure to the S&P 500 up and down with the market regime. It is re-derived every night by an autonomous research loop and executed through Alpaca. No discretion, no overrides.
The read
After every close, nightclaude reads the day's tape (trend, breadth, and how violently prices are moving) and turns it into a single decision: how much conviction to carry into the next session.
When the evidence lines up behind a calm, trending market, it leans in. When volatility spikes or participation narrows, it scales back toward safety. Exposure slides continuously between fully defensive and maximum conviction. It is never an all-or-nothing bet, and never a discretionary one.
How it stays honest
The hard part of systematic trading isn't finding a rule that fit the past. Anything fits the past if you torture it enough. It's telling a real edge apart from a lucky one. nightclaude is refined by an autonomous research loop adapted from Karpathy's autoresearch pattern: propose a change, test it across many windows it was never fit to, and keep it only if it survives.
Every candidate is judged with the toolkit quant researchers use to fight self-deception: walk-forward validation, the deflated Sharpe ratio (which discounts for how many ideas were tried), and the probability of backtest overfitting (PBO). Most ideas die here. The survivors are the ones that held up on data they were never fit to.
Execution
Each night the strategy computes the next day's target. The order is placed shortly after the following morning's open, through Alpaca, the commission-free brokerage API.
Alpaca is the source of truth.Current value, positions, and the equity curve are read live from the broker. The site keeps no copy of its own. The only things stored are each night's target and the link from every fill back to the decision behind it.
Honest disclaimers
Backtests are not reality. Orders fill at the next morning's open, not a backtest's same-day close, roughly sixteen hours apart. Overnight gaps and slippage will pull live results away from any backtest.
The strategy will have bad days. A leveraged, volatility-aware strategy takes double-digit drawdowns; expect occasional sharp losses. Don't panic on a red week.
Not investment advice. This is a transparency project, not a recommendation. Past performance does not predict future results.