The numbers that move with your lot: seven that travel and three that do not
Written on 4 September 2026. Counts, rankings, extremes and labels here describe the audits on record that day; audits published or rebuilt since have changed some of them, and retired audits are not current evidence. Since 29 September spread sensitivity per trade is tagged sizing dependent, because it is a USD amount per trade; the audits now list six invariant metrics and four dependent ones.
Every backtest in this catalog was run from one position size input, a 0.01 base lot, and nobody reading it will trade that size by accident. Some EAs still size their own trades on top of it; each audit states the lot policy it measured. So each audit sorts its own metrics into two groups and prints the result. Seven of them keep their value when the position size changes. Three of them do not, and those three are the ones a product page likes to quote. The legend sits in the box headed What this audit measures, directly under the line Not measured here, and 22 of our 26 default audits carry it.
The two tags, word for word
The seven tagged sizing invariant are Profit concentration, Active trading days, Trade count, Spread sensitivity per trade, Drawdown duration in trading days, Holding times and Win rate.
The three tagged sizing dependent are CAGR, Drawdown in percent and Absolute PnL.
The split is declared once in the engine and then travels inside the audit file, so it reaches anyone who downloads the JSON instead of reading the page. It is not a ranking of which metrics matter. It is a statement about which ones stop being true the moment the size changes.
Why it has to be printed at all
Our measurement frame is deliberately large and deliberately boring. Every default run uses a 0.01 base lot input on a 100,000 USD account at 1:5000 leverage, and every audit repeats those three values in the same box. A frame that big stops a losing streak from ending the test early, which would turn a strategy result into a margin call.
The price of that choice is that every money figure and every capital ratio in the file belongs to the frame rather than to the reader. A trade count belongs to the strategy. A profit in USD belongs to the account it was earned on. Without the legend those two kinds of number look identical, because they are printed in the same table.
One run, both kinds of number
Gold Snap traded gold on M1 from 2003.05.05 to 2026.08.16. The audited net after our commission retrofit is 1,922.00 USD, the end of day series puts compound annual growth at 0.08 percent and the deepest end of day drawdown at 0.54 percent, both measured against the 100,000 USD frame. Those are the tagged figures, and every one of them would have to be recomputed for any other size.
The same file, the same run, read on the other side of the legend. The run made 3,769 trades. It was active on 2,603 of the 6,075 trading days in the window, a share of 42.85 percent. It needed 169 of those active days to earn 80 percent of its profit. The average trade lived 1,550 seconds. None of those five figures moves when the lot moves, because none of them is measured in money.
The same reading on a larger result
Gold Trade Pro ran on the daily timeframe from 2003.05.05 to 2026.08.22 and finished with an audited net of 13,446.15 USD, compound annual growth of 0.53 percent and a deepest end of day drawdown of 1.38 percent on that same frame. Its invariant side reads 6,305 trades, 1,805 active trading days, 71 days for 80 percent of the profit and an average holding time of 25,402 seconds.
Set the two runs side by side and the legend earns its space. The money lines describe an account nobody will open, because nobody deposits 100,000 USD in order to trade 0.01 lots. The counts, the days and the holding times describe the same behaviour at any size, and they are the part of an audit that survives being carried into a real account.
What the audit will not do for you
The legend says which figures have to be recomputed. It does not recompute them, and it does not predict what a larger size would have produced. That distinction is the whole point. A tag that promised the result scales would be a forecast, and a backtest cannot carry one.
Four of the 26 default audit files do not carry the block at all. They were built by an earlier engine version, before the tags existed, and until they are rerun in our own test setup their metrics stay untagged. Adding a legend to them now would describe a frame that was never ours.
Run these checks yourself
- Find the deposit before you read the percentage. Every percent return and every drawdown percentage in a backtest is a ratio against a deposit, and the deposit is a choice somebody made.
- Carry the counts, not the currency. Win rate, trade count, holding time and active days survive a change of position size. A profit in USD does not.
- Ask what a headline number is a ratio of. One trade list produces very different percentages on different account sizes, and the trade list is the part that was actually tested.
- Treat a missing frame as missing information. A result published without its deposit, its lot size and its leverage cannot be checked by anybody.
The legend, the frame and every figure quoted above are public in each audit we publish. Your own tester report? The browser check is free.
The honest limits. The legend covers pure
scaling of a fixed lot and nothing else. The same box states that
individual risk sizing is not measured here, so an EA whose own money
management grows the lot as the balance grows produces a different
trade list rather than a scaled one, and no tag applies to that case.
A larger size also changes the margin the account needs, and the tags
say nothing about margin. The three dependent metrics are not wrong.
They are correct for the frame they name, and the frame is printed
next to them for exactly that reason.