Lizard does not blow up. That has to be said first, because the failure this catalog is built to catch is exactly that one. It is also not the common outcome. Across our 25 audits the deepest reconstructed balance drawdown of any run is 37.2%, and 22 of the 25 finish the tested history in profit. Lizard is one of them, and its problem is a different one. Over 23 years of gold ticks it opened 21,524 trades across 4,242 trading days, held them for an average of 227 seconds, never once carried a position overnight, and finished with a maximum drawdown of 2.55%. The deepest hole it ever dug on a $100,000 account was $2,545.79.
It also finished the 23 years with $183.99 of profit.
The cost block is the whole story and it is short. Gross before costs was $1,905.91. Our commission retrofit at $3.50 per lot per side charged 43,048 deals and came to -$1,721.92. Swap was $0.00, because 0 trades were held overnight. Final balance $100,183.99, reconstructed from the deals and matching the report to the cent.
Costs took 90.3% of gross profit. Per trade that is under nine cents earned and eight cents paid. What survives is less than one cent per trade, 21,524 times over.
The leave best out table says the same thing from the other side. Remove the 10 most profitable trades out of 21,524 and the run ends at $7.17. Remove 20 and it ends at -$130.08. Twenty trades, one in a thousand, decide whether this is a winner.
Lizard was published on 2026-05-09. Our release split cuts the same run at that date. Everything before the release, 8,398 days and 21,361 trades, comes to -$252.61 at a profit factor of 0.97 and a win rate of 29.5%. The 96 days from release to the last vendor update inside the tested window, 163 trades, come to +$436.60 at a profit factor of 6.73 and a win rate of 83.4%.
Twenty three years of gold history produced a small loss. Three months produced everything the run shows. Those three months are also the months in which the vendor shipped 23 updates to this EA. We are not calling that fraud. We are pointing out that the profitable part of this backtest is exactly the part that was visible while the current version was being built, and that the genuinely out of sample window, the slice after the last update, is 0 days long in this run.
The year table puts a number on the same effect. 2026 carries 445% of total profit. Ten of the tested years are negative. Take 2026 out and there is nothing left to discuss.
The shallow drawdown deserves a second look, because shallow is not the same as short. Our deepest episode ran from a peak on 2003-05-13 to a trough on 2007-11-08 and did not recover until 2026-06-18. That is 6,027 trading days underwater out of 6,067 in the end of day series.
Read that as the honest version of the smooth equity curve. An owner who started this EA at the beginning of the tested history spent essentially the entire history below their opening balance, by a small amount, waiting. The maximum drawdown of 2.55% is true and it is not the number that would have decided whether they kept the EA running. The 23 years of waiting is.
Same instrument, same lot size, no strategy. Gold moved 1,161.43% across the tested span, from 350.488 to 4,421.15. Holding 0.01 lots through it would have made $4,070.66. Lizard made $183.99 on the same size, which is about 22 times less, while trading 21,524 times and paying commission on every one of them.
That comparison is unfair in one direction and worth making anyway. A buy and hold position carries risk this EA never takes, and the EA was flat overnight and over every weekend. But the buy and hold figure is the bar any long biased gold system has to clear, and this one is nowhere near it. 68.3% of what profit there is came from the long side.
Our withdrawal replay sweeps everything above the start balance to the owner on the last trading day of each month. Across 280 months it paid out in 3 of them, $185.00 in total, which is 0.01% per year. The longest dry streak is 277 months.
Three paydays in 23 years. That is not a drawdown problem, it is the absence of an edge that survives its own costs.
The MQL5 listing recommends a $250 deposit. We replay the observed trade path from every one of the 43,048 deal timestamps as a fresh start on that deposit and ask how many of those starts hit zero. The answer is 61.0%.
The same trades that produce a 2.55% drawdown on $100,000 kill six out of ten small accounts, because on $250 the same dollar swing is a different fraction. This is the recurring gap between a measurement frame and a real account, and it is why our audits carry both.
The sweep is not about death here, and in this catalog that is the normal case rather than a distinction. 21 of our 25 audits show zero floor breaches at every account size we test. Against IQ Capital Classic the account survives at full size and at every reduced size we test, with 0 floor breaches and 0 per position loss breaches throughout. The modelled withdrawal rate at full size is 0.01% per year, and at half size and below it rounds to 0.00%.
What fails is the consistency rule, which caps how much of a year's profit a single day may carry. It is breached in 6 years. Concentration explains why. 80% of the profit was made in 3 of the 4,242 trading days. We say plainly that when the total is $183.99, a statement about where 80% of it came from is arithmetic as much as behaviour. The rule does not care about that distinction, and neither does a funded account.
This is not an averaging grid, and the audit says so. Volume never rises after a loss, 0.0% of the time, and the median size of an adverse add is 1.0 times the basket average. What it does do is stack. 9,191 entries were opened on top of existing positions, 42.7% of all entries, up to 10 positions at once, of which 3,827 are split tickets from one signal. 4,572 adds went with the move and 695 against it. Long and short were open together in 13 moments. Our structure verdict is caution rather than red for exactly that reason. The exposure scales with the stack, the geometry does not compound it.
One number in the exit mix is worth naming. Of 8,206 exits on a stop loss, 5,623 closed in profit, which is 68.5%. That is the shape of a stop that follows the price rather than one that caps a loss. Our trailing detector did not fire on the ticket pattern, so we report the count and leave the mechanism to the vendor's documentation.
The measurement is clean and the audit should say so. 100% real ticks from 2003.05.05, 1,409 absent minutes out of 7,897,519 minute bars, 0 randomizer prints, so the run is deterministic and another tester can reproduce it. 0 failed entries. Chronology violations 0, duplicate deal ids 0, log coverage 1.0, and the reconstructed final balance matches the report exactly. Data quality is the one dimension we rate ok without qualification.
Two head figures belong to the tester, not to our cost ladder, and we name them as such. The report head prints a profit factor of 1.21 and a Sharpe of 11.81, both before our commission retrofit. Our own end of day Sharpe on the same balance curve is 0.09, and the compound growth rate is 0.01% per year. A Sharpe above 11 is what you get when you divide a tiny mean by a tiny per trade variance. The mean itself is $0.0085 per trade, below the $0.01 of extra cost per trade that flattens the entire run.
This audit measures v1.87 from 2026-08-12. On 2026-08-18 the vendor published v1.88, which reworks the session filter, blocks deselected sessions completely, extends the error log and removes one preset. None of that is in the version we measured, and none of it changes the cost arithmetic in an obvious direction. Fewer trading sessions means fewer trades, which cuts commission and the edge together. Whether the ratio improves is an empirical question, and we track the update for a re run rather than guessing at it.
The deal level census, the release split and the fresh start survival map need deal reconstruction, and that is what we sell. Any catalog audit is $19, and EAs with several audited presets or versions come as one bundle. Your own tester report? The browser check is free.