One run, two capitals: which audit numbers move when the account does

2026-09-03, measured across every default audit in the catalog that carries the capital blocks. The figures below sit on the full audits; those of EAs listed or rented at 200 USD or less are free to read.

Figures checked on 3 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.

Every run in this catalog is measured on a 100,000 USD account, and not one of them is judged on it. The tester deposit is a measurement frame. The verdict is computed on the deposit the vendor recommends, and across the 22 default audits that carry the block that figure runs from 100 USD to 2,000 USD. Same trades, same window, two capitals, and only one of them belongs in a percentage.

Why the frame is large

A small account changes the test itself. If a losing streak eats the margin, the tester stops opening positions and the run ends early, so the result describes a margin call rather than a strategy. The frame is therefore set high on purpose, 100,000 USD with a 0.01 base lot input in all 22 audits, and the evaluation moves to the vendor's own number. Both values live in the same file, run capital and evaluation deposit next to each other, so neither has to be remembered while reading.

What makes this legitimate rather than convenient is the fixed lot. With a lot size that never changes, the future profit and loss in USD from any moment on is identical for every account size. The frame therefore does not move a single trade. It only changes what the trades mean.

The same trades on two accounts

Scalping Robot Pro is the clearest case. On the frame the run ends at 46,765.32 USD after our commission retrofit, a loss of 53,234.68 USD and an account that survives to the last day of the window. On the 1,000 USD deposit the audit evaluates against, the same trade list kills the account 53 times over, and 97.0 percent of the 605,596 possible start points end dead.

Lizard tells the same story more quietly. Its run ends at 100,183.99 USD on the frame, which is a profit and a survival. Started with the 250 USD the vendor recommends, the identical chain dies 10 times, and 61.0 percent of 43,048 start points do not make it.

Three further audits carry a death share above zero, Logan at 10.2 percent with 10 deaths on 2,000 USD, Gold Snap at 14.6 percent with 2 deaths on 200 USD and Smart Gold Hunter at 12.7 percent with 1 death on 1,000 USD. The other 15 of the 20 maps never die at all, and that is the honest half of the finding. A frame that flatters is not the same thing as a frame that lies, and most of these runs read the same way on both capitals.

How a start point map is built

The map asks one question at every point of the balance curve. If somebody had opened an account with the vendor deposit on that day, would it have survived the rest of the history? With fixed lots the answer needs no second backtest, because the future result in USD is the same for every account size. An account dies when the chain of future results falls at least the deposit below the level at its own start, which is a single scan backwards over the curve. The death chain applies the same idea once from the beginning and restarts the account at every death, which is where the counted deaths come from.

Which numbers survive a change of size

Every audit ships a small table that says which of its own figures are size invariant. Six are, profit concentration, active trading days, trade count, drawdown duration in trading days, holding times and win rate. Four are not, CAGR, drawdown in percent, absolute profit and loss, and spread sensitivity per trade, which is a USD amount per trade and grows with the lot. The split is declared by us and pinned by an engine test rather than measured per run, and it follows from the arithmetic. Pure lot scaling changes money and capital ratios and leaves the shape and the timing of the trades untouched.

The practical consequence is short. When two products in this catalog are held against each other, the six invariant figures compare as printed. The other four carry their basis with them and compare only against the same basis.

Six deposits we chose ourselves

The evaluation deposit is a vendor claim and not our measurement. In 16 of the 22 audits it comes from the product page. In the remaining 6 the vendor names no deposit at all, the audit falls back to 1,000 USD, and the file says so in words. Gold Atlas, Prop Firm Gold EA, Range Breakout EA with Range Filters, Smart Gold Hunter, Smart Gold Impulse and The Gold Reaper carried that label on 6 September. A number we picked is a weaker basis than a number the vendor published, so it is marked instead of blended in. (Since then Gold Atlas and The Gold Reaper evaluate on the vendor's own figures, 1,000 and 600 USD; their audit pages show the source.)

Run these checks yourself

Every audited EA carries both capitals on its audit page with the deal level evidence behind them. Your own tester report? The browser check is free.

The honest limits. The start point map needs fixed lots. Two of the 22, Gold Trade Pro and The Gold Reaper, carry no map at all because their measured lot policy came out as mixed or unknown and the gate stays closed rather than guessing, which is why the death figures above count 20 audits and not 22. Four default audits carry neither capital block, because an older engine version built them, and they are absent from the 22 rather than counted as zero. The death criterion is the latest possible one, an account that has lost its whole deposit, so a real broker closing positions earlier would kill some of these accounts sooner and the counted deaths are a lower bound. The evaluation deposit is a vendor figure captured on the day the audit was built and vendors change it. And none of this says that a small deposit is wrong. It says the same run reads differently on each capital, so the audit prints both instead of picking the flattering one.

More from the blog

Nine of twenty earned less than they cost: list price against measured net profit (September 2026)

Lizard closed 21,524 trades on gold and finished 183.99 USD ahead; its licence cost 199 USD when its audit last checked the price (31 August 2026). Twenty of our twenty six default audits carry a list price next to a measured net result, and nine of them earned less across the whole test window than the software costs to buy. One 299 USD purchase of Smart Gold Hunter contains four presets whose results lie 3,467.13 USD apart.