Strategies as Data, Not Code

I entered a Jupiter Developer Platform bounty with a project called jup-policy. The pitch: most trading bots hardcode the strategy, so every new rule is a code change and every backtest is a git branch. I inverted it — the strategy is a JSON file. You declare when conditions (24-hour price moves, token health stats like organic score and holder count) and then actions (limit orders, DCA schedules), and a generic engine polls the price APIs, evaluates your rules, and emits the orders.

The part I’m most careful about: by default it never signs or submits anything. It builds the unsigned transaction, hands it to you, and stops. After writing about how online gambling gives you a small edge that isn’t big enough to live off, I wasn’t about to build a machine that YOLOs my actual money — the interesting problem was the policy engine, not the position.

Constraints expressed as data instead of code. The same idea keeps showing up everywhere I look, from BI dashboards to this thing.