Quantum Titan review: three versions in two weeks, measured

Quantum Titan was published on the MQL5 market on 2026-08-07 at 1,199.99 USD. By 2026-08-21 it had received two updates. We audited every version we could install, v2.1, v2.2 and v3.0, in the same measured world: 23 years of real gold ticks, identical frame, identical costs. The version story is real. Version 3.0 nets 1,089.53 USD where v2.2 netted 488.07, the end of day Sharpe rises from 1.55 to 1.87, and the two consistency breaches that held the prop lamp at caution in both 2.x versions disappear: the propfit lamp turns ok. The other finding is just as hard: of 1,947 trades in the v3.0 run, 1,940 lie before the release date. There is no out of sample window at all.

What we ran

Standard catalog protocol. XAUUSD tick history from 2003.05.05 to 2026.08.23 on M1, every tick based on real ticks, 10 ms execution delay, fixed 0.01 base lot on the 100,000 USD measurement frame at 1:5000, capital math on the vendor recommended 1,000 USD. The journal pairs all 1,947 trades exactly. Costs after our commission retrofit of 3.50 USD per lot per side take 13.7% of gross profit (commission 155.76 USD, swap 16.61 USD, 61 overnight trades, weekend exposure avoided by default).

What actually changed between the versions

The matrix on the audit page compares the three runs in one world. From 2.1 to 2.2 the backtest got worse (net 669.83 down to 488.07, Sharpe 1.75 down to 1.55). From 2.2 to 3.0 it more than doubled, and the profit factor in the report head reads 1.5 on 1,947 trades. The structure stays modest throughout: at most 2 simultaneous positions, 254 stacked entries of which 226 add with the move, no martingale signature, a recovery mode that exists as an input but ships disabled. Two sub strategies share the work, the main one holding for a median of 0.07 hours, about four minutes.

Read that sequence honestly. A vendor iterating three builds in two weeks against a visible 23 year history has every opportunity to fit, and the only slice the release calendar leaves as a test is the 7 days between release and the first measurable update: 7 trades, net -4.81 USD. The window after the last update is zero days long. Nothing in this data can distinguish a genuinely improved engine from a better fit to the same past.

The lamps that stay on

Two cautions survive into v3.0. The log records 1,260 failed entry attempts, our data quality dimension flags that. And 80% of the profit concentrates into 55 trading days (4.4% of 1,256), the familiar calendar problem of gold systems. The header prints a Sharpe of 9.25 and a relative drawdown of 0.14% on the deep frame. Our end of day recomputation says 1.87, which is good and roughly one fifth of the marketing number. The IQ Capital replay at full tested size shows zero deaths, zero position limit breaches and a withdrawal line of 16% a year on the vendor deposit basis.

Run these checks yourself

• Date every version against the test window. A product younger than a month cannot have out of sample evidence. Treat all of it as in-sample until the calendar says otherwise.
• Ask what an update changed, in numbers. Vendors write changelogs, backtests write deltas. If nobody measured the versions in the same world, the changelog is marketing.
• Check the failed entries line in your own tester log. 1,260 rejected orders in a backtest are a mechanic worth understanding before real spreads and real rejections arrive.

The full version matrix with per version lamps, the consistency tables and the sizing sweep are in the catalog audit (one bundle for all three versions). Your own tester report goes through the same engine: the browser check is free, the full audit is $19.

Related: the audit page for this EA · all audits · more articles