How it works

What the algorithm looks at

It reads 15 minute candles of 13 CME Group futures: the four US index futures, gold, silver, copper, crude oil, natural gas, gasoline, corn, wheat and soybeans. It uses only candles that have fully closed. It never sees the current, unfinished candle.

The approach is market structure, sometimes called Smart Money Concepts. In short: price makes swing highs and swing lows. When a candle body closes beyond a recent swing, structure has broken in that direction. The algorithm then looks back along the move that caused the break for the last candle that went the other way, the order block, and places a limit entry inside it with a stop just beyond it. Targets sit at one, two and three times the risk.

Optional confirmations such as a liquidity sweep, a fair value gap and the one hour trend are recorded as tags on every trade. In the first version they are not required. That is deliberate: record everything first, then tighten with evidence.

How a trade is graded

• A call is pending when its signal candle closes. If price does not touch the entry within 8 candles, it expires and does not count in the win rate.
• Once filled it is open. If the first target is touched before the stop, it is a win. If the stop is touched first, a loss.
• If one candle touches both the first target and the stop, it is counted as a loss. The order inside a candle cannot be known, so the conservative answer is used.
• If neither is touched within 32 candles after the fill, it is a timeout and the result at that close is recorded.
• An opposite call on the same symbol cancels a pending one. Cancelled calls are shown but not counted.
• Wins keep being tracked, with the original stop, to record whether the second and third targets were also reached.

Candles are counted, not hours. Across the daily maintenance break, a weekend or a thin session, 8 candles is longer than 2 hours.

Win rate is wins divided by wins, losses and timeouts. Net R uses the simplest model: every win is +1R, every loss is -1R, a timeout is its recorded R. Commissions and slippage are not modelled, so the real break-even win rate is above 50%.

Why everything is one hour late

Trades appear 1 hour after the algorithm calls them. These are not signals. The delay is enforced by the database itself: the public part of the site can only read a view of the data that is at least one hour old, including outcomes. A grade that landed five minutes ago still shows as open. Nothing on this site is a signal to act on.

Versions

Every change to the rules or their numbers ships as a new engine version. A trade is forever tied to the version that called it and graded under that version's rules. Statistics are kept per version, so when something improves you can see whether it did, instead of being told it did. Recorded trades are never edited or deleted; the database refuses.

Honest limits

• Results are simulated. No orders are placed. Fills are assumed when a candle's range touches the entry.
• The data feed is a free, delayed, continuous-contract feed. Contract rolls can create price gaps; trades affected by a large gap are tagged.
• The sample is small at first. Months, not weeks, are needed before any number here means much.
• No machine learning or language model is involved anywhere in the decisions. The engine is plain, deterministic code: the same candles always produce the same calls.

The full disclaimer applies to every page.