Crucible reconstructs your system against order-level ES market data — real slippage, real queue position — and returns a verdict judged on return per unit of drawdown. Survival, not a flattering equity curve or a borrowed Sharpe ratio.
Fitted to history, blind to fills. It flatters the curve and says nothing about what happens when the book moves against you.
Reconstructed at order level under real slippage and queue position, then judged on the only thing that funds a live account: surviving the drawdown.
Entry and exit logic, parameters, stop and target, the window you want tested. No data wrangling or infrastructure on your side.
You don't supply — or pay for — the market data. Crucible licenses the order-level ES feed; that expense sits with us, not you.
Your system runs against market-by-order ES data with realistic slippage and queue position — fills as the book would actually have given them, not mid-price fantasy.
Return per unit of drawdown, loss clustering across shock regimes, calibration, and live/backtest parity. The measures that decide whether an edge lasts.
Any ML model is trained on at most four years of data. Older history carries a different market microstructure and regime — it dilutes more than it informs, so it's deliberately left out.
A written report that ends in GO or NO-GO, with the reasoning laid out. No dashboard to interpret — a decision you can stand behind.
Scope at launch: ES futures, intraday systems. Founding pricing is limited and ends once the cohort is full. Other instruments by arrangement.
No. You share the strategy logic — entry and exit rules, parameters, stop and target structure — in whatever form is clearest. I don't need your implementation, and I don't redistribute or reuse anything you send.
On market-by-order (MBO) ES futures data — the most granular source there is. If your model was built on L1 or L2, I reconstruct that exact level faithfully from the MBO source, so you're validated at the granularity your model actually trades on — then I show you the slippage that level was hiding.
Most backtesters assume optimistic fills and judge a strategy on its equity curve or Sharpe. Crucible reconstructs realistic fills from the order book and judges on return per unit of drawdown — survival. The question isn't "did it look good," it's "does the edge hold when execution and drawdown are real."
ES futures, intraday systems, at launch. Other instruments are possible by arrangement — ask.
A written report that ends in a clear GO or NO-GO, with the reasoning and the diagnostics behind it. Not a dashboard to interpret — a decision you can act on.
Five business days for a Baseline validation. Priority turnaround is available on the retainer.
Then you've saved an account. A NO-GO comes with the reasons the edge didn't survive — which is often more actionable than a GO, because it tells you where the strategy breaks.
Send the strategy and the window you want tested. You'll get a scope confirmation and a turnaround date — not a sales call.