quantdesk

Account equity

Balance after every deal that has closed, in the order they closed. There is no time axis, because four round turns cannot support one: each step is a trade, not a day. The commission probe is in the path and drawn in amber, because it moved the balance for real even though it tested nothing. The last segment is dashed: that is the open book, and it is not settled yet.

Open positions

Every row is money already at risk. Risk is the stop distance times the value of a price point times the lots, which is what the account actually loses if the stop fills. Reward / risk is only printed where a target price exists. Tap a row for the full arithmetic and the exit rule.

Per strategy

Realised, from closed round turns. Payoff is the average winner over the average loser after cost, which for a strategy with no take profit is the only honest risk-to-reward figure that exists: it is measured rather than intended. Tap a row for what the strategy is and how it is sized.

Net by strategy

Zero is the line down the middle, so a losing strategy reads as losing rather than as a short bar. Tap a bar for the detail.

Closed trades

Round turns only. Entry fills are excluded, so the count is trades rather than executions, and net is after commission and swap.

Expert log

Tailed live. This is where an EA that thinks it is running but cannot trade says so, and where a rejected order gives its reason.

loading…

The finding

loading…

The 15 cells running live

Each card is one row of the board, running on the demo exactly as it was measured. The four bars are net after cost in each measurement window, oldest on the left. A cell that is only green in the window it was picked from has told you something. Tap a card for its full provenance: sizing, exit rule, every window, and what its stop costs.

Why holding longer is the only thing that helps

Three quantities on one hold-time axis, all measured on the same 15,996 cells. Lift is what the trigger adds over an entry at the same minute of day on a random date. Cost is the round-turn toll in the same noise units. Power is the effective sample size the t-statistic is actually built on, which is bounded by trading days rather than by entries.

Window

Lift (top 5% of cells) Round-turn cost Power, effective √n

The bar every result has to clear

Search N rules on pure noise and the best one comes back with a t-statistic of about √(2 ln N) for free. At N = 15,996 that bar sits at 4.40. The only honest way to show a leaderboard is to draw the bar on the same axis.

What the stop costs

Measured per live cell on two years of paths. The mean without a stop column has a known correct answer of zero, because a random entry with no cost and no stop cannot have an expectation. It reads zero on every row, which is how this engine gets checked instead of trusted. The stop then costs almost nothing in expectation while cutting the worst case by three to six times, and that is the whole argument for never holding a position without one.

What the triggers add

Lift across every rule cell. If the triggers carried information, this would sit to the right of zero. Instead the median is negative in all four windows and slightly under half the cells are positive, which is what a coin flip minus a small cost looks like.

Level families

Median lift per family, over every cell in it. The median is the right summary for asking whether a family works and the wrong one for finding the single cell that does, so both are on the page. Tap a bar for the family's numbers.

The 45 that clear the bar on lift

These are real triggers. Every one is at 15, 30 or 60 minutes, and every one is priced out: the toll runs 5 to 10 times the edge. Not one of them clears the bar once cost is subtracted.

Full board

15,996 cells: 16 instruments x 25 strategy families x 10 hold times x 4 time windows. t lift is the t on what the trigger adds, which the board does not carry natively; it is computed generously, so a cell that fails it is dead without needing the exact version.