Plain English in. A quant's contract out.

Your sentence is compiled once into a deterministic contract: a versioned document of rules, signals and risk limits. Backtesting and paper trading run only that contract, never a model’s judgment. What you approve is exactly what runs.

The compiler asks a question rather than guessing a number you did not say, and when a strategy cannot be tested honestly it refuses and says why.

Every backtest charges spread and slippage on every trade, so every result is net of the spread and slippage the cost model charges. It reads data only as it was known at the moment of each decision: a price from later that day, or a report filed the following week, cannot leak into a trade made before it existed.

Every run is pinned to one sealed copy of the market data, and every result says which window it covers, where the data behind it begins and what the cost model assumed. Run the same contract on the same copy and you get the same number, and so does anyone else. Thin samples, unstable windows and thresholds that flip the result get flagged in the margin. Skepticism is a product feature.

Publishing a result re-runs it on our servers: what gets stored is what our servers computed, not what a browser showed. Every shared report says which day its data was sealed and carries a re-run button that runs the same contract on that same data, for as long as we hold it. If the number comes out different on the same data, the page says so, because that is a fault of ours and not the market moving; if we no longer hold that data, it says that instead of pretending.

The Leaderboard ranks strategies by Sharpe over the full window: return per unit of risk, across every year the strategy was tested, costs included.

Every backtest runs on this, and no further back. A strategy over an index starts where our membership for that index starts, and its result says so.

Anyone can create an account. Nothing is billed.