The magic, made explicit

The two gates

Nothing we sell is sold on backtest evidence.

A strategy that looks good in simulation has proven nothing — most do, then die on contact with a live market. So every strategy our Central Agentic Intelligence, Eli Halden, designs must clear two gates. Together, those two gates are what we call the Gauntlet: a discovery half, where no real money is at risk, and a live half, where it is. Only what survives the second is ever allowed to become a product.

The build is the magic; the machines are the product.

Gate 1

Discovery — no real money at risk

Where the edge is found or disproven. A fixed sequence, run before a single real dollar is at risk:

1

Hypothesis

A single, specific, testable edge, stated in one sentence. Pre-registered before we look at the result. No black boxes.

2

AI design & code

Eli Halden designs the strategy and writes it as an expert advisor.

3

Historical backtest

Run across years of market data to see whether the edge ever really existed.

4

Out-of-sample validation

Re-tested on data it never saw during design, with the sign required to stay stable in-sample → out-of-sample.

5

Real broker-cost gate

Spreads and commissions modeled from real broker data. It must survive real costs, not the idealized fills that make most backtests lie.

6

Regime testing

It has to hold up across market conditions, not just the one it was born in.

7

Family false-discovery control

The whole candidate family is corrected together, so no single winner is just luck surfacing from many attempts.

Three outcomes

Gate 2 · the ship gate

Live forward verification

This is the gate that decides what you can buy. Anything that survives Gate 1 — validated or provisional — goes onto a small live, real-money account. Third-party tracked. Public from day one. And the kill rule is written before it starts — minimum number of trades, maximum drawdown, and an expectancy floor — so we can't move the goalposts once it's running.

Two outcomes

Family discipline. Every candidate we test counts against a shared multiple-testing budget, and every dead hypothesis is retired from the family so it stops consuming statistical power. The more we throw away, the harder the next candidate has to work to look real — the opposite of fishing for a winner.

Some questions we refuse to fake an answer to. Week-long strategies sound testable — until you count the data. Twenty years of weekly charts is only about a thousand candles, and a thousand flips can't tell you whether a coin is slightly bent. Run the test anyway and you get a shrug dressed up as an answer. So we don't run it. That question waits until we find a way of asking it that can actually hear the answer — and until then we say “untested” out loud instead of pretending.

What we don't do: no grids, no martingales, no doubling-down, no “guaranteed” anything. Single-strategy systems with defined risk. If we can't ship a product honestly, we don't ship it.

Trading involves substantial risk of loss and is not suitable for every investor. Backtested and out-of-sample results are hypothetical, have inherent limitations, and are not a guarantee of future performance. Live verification results, where shown, are a real but limited track record. Past performance is not indicative of future results.