Get started in 30 minutes

Understand the core idea (2 minutes)

ConvexPi has two things:

• The Lab — write a Python strategy, submit it, and the grader measures how well it generalizes to hidden data. Your score is the OOS Sharpe ratio: positive means real alpha, negative means noise.
• The Arena — write a trading agent that places orders on a live limit-order book. Compete against other agents in real time.

Start with the Lab. The Arena comes later.

Run Mission 1 in Colab (15 minutes)

Mission 1 is the overfitting game. You will build a strategy, see it score well in-sample, then watch it fail out-of-sample. Then you will learn why, and fix it.

No local setup needed. Colab runs in your browser. pip install convexpi is the only install.

Create an account

You need an account to submit strategies and appear on the leaderboard.

Already have one? Log in

Submit your strategy (5 minutes)

The demo competition is always open. Paste the strategy you built in Mission 1 into the editor and submit. The grader runs in under 5 minutes.

from convexpi.lab import Strategy
import numpy as np

class MyStrategy(Strategy):
    def on_day(self, day, features, prices, portfolio):
        sig = features.get('mom_1m', np.zeros(len(prices)))
        sig = np.nan_to_num(sig)
        total = np.abs(sig).sum()
        return sig / total if total > 0 else np.zeros(len(prices))

See your score on the leaderboard

After grading completes, your OOS Sharpe appears on the public leaderboard ranked against all other submissions. Three reference baselines are always pinned so you know what "beating random noise" looks like.

Your grade report (IS Sharpe, OOS Sharpe, overfitting ratio, alpha discovery) also appears in your submission history on the submit page.