1Fundamentals

The AI Pro100 family robots are a tool for high-risk margin trading designed to grow an account fast. Xaurus AI Pro100 and Swinger AI Pro100 are two different trading systems (trend-following and counter-trend) with an identical AI Pro100 self-optimization mechanism: their inputs, controls and this guide are shared, so everything written here applies equally to each of them. The trading logic differs between the systems; it is not disclosed and is not configured directly.

1.1Risk profile and the dedicated-funds rule

A trader using AI Pro100 always risks 100 % of the funds placed on the trading deposit. Money that must be protected from loss should be kept out of this trading — never deposited. A sensible approach is to run the high-risk trading in parallel with conservative instruments (bank deposits, real estate) holding the rest of your capital, optimizing the overall risk profile of the portfolio.

1.2Trade series and virtual levels

The EAs place no pending orders and no broker-side stop-losses or take-profits (the only exception is the Equity SL % protective stop, §1.3). Instead, all levels are “virtual”: the EA draws them on the chart and executes them itself. Levels are placed, moved and executed no more than once per minute bar. This is deliberate behaviour, tied to the architecture of the trading as a whole.

The absence of an SL in the order properties does not mean the EA “trades without stops” — it virtualizes them, among other reasons because we are convinced the broker has no need to know our trading levels before the moment they are actually executed.

1.3Drawdown control: Equity SL %

The Equity SL % input sets a percentage of the account balance taken at the start of the current trade series, below which the account equity must not fall while that series develops. As long as series end normally — at a profit — the reference balance in practice equals the latest balance high-water mark (an emergency series ending — the protective stop, Stop out — halts the EA, §8.5).

The mechanism is enforced by a real broker-side stop-loss: the EA computes the common price level at which account equity would touch the threshold and places it as the stop-loss on all of its open positions (recomputing on every new opening inside the series). When this stop fires, all positions are closed and the EA stops trading — see §8.5. A value of 0 disables the mechanism.

1.4Profit target: Equity TP $

The Equity TP $ input is the target account equity in the deposit currency. As soon as equity reaches this value while the EA is active, all positions are immediately closed at market and the EA stops trading. A value of 0 disables the mechanism.

Set the target with a margin for closing slippage — for example +10 % over the goal if the aim is to secure at least a given percentage. Remember that +200 % profit corresponds to multiplying the initial deposit by 3 (not 2). Example: starting deposit 7,000, target +200 %, 10 % slippage margin — Equity TP $ = 7000 × 3 + 10 % = 21,000 + 2,100 = 23,100.

We recommend deriving the profit target itself from the set's test period: plan for roughly ¼ of the historical span on which the set (SSN) tested profitably. Example: a set optimized on the last 2 years with an average profit of 20 % per month — plan for 6 months of work (2 years / 4) with a result of 1.2⁶ ≈ 3, i.e. about +200 % in half a year (an example of target arithmetic, not a return forecast). Once the profit or time goal is reached — re-optimize for the latest market (§5) and start a new round: AI Pro100 trading is built on efficient self-optimization with minimal trader involvement.

1.5The SSN code and closed settings

The EA does not disclose its trading parameters. That is not only protection against copying and reverse engineering — it is a usage methodology: the trader gets by entirely without deep knowledge of the internals while keeping the full power of flexible optimization for any market — on equal terms with the developers. In our own trading, including our public signals (copy service), we use exactly the same mechanism and the same EA version that we distribute.

We publish up-to-date sample sets in open sources — in particular, they can always be found on the product page at mql5.com. The core idea of AI Pro100, however, is different: every trader — even before purchasing the robot — can use this guide to run the powerful self-optimization engine and find settings that suit them personally: their market, broker, capital and risk appetite. The SSNs we publish are examples of what the engine produces, meant to demonstrate its capabilities; they are not trading recommendations.

1.6How AI Pro100 optimizes

The AI Pro100 methodology applies artificial-intelligence techniques to multi-parameter optimization: the search covers a settings space that brute force could never traverse in reasonable time. Optimization runs in the standard MetaTrader 5 Strategy Tester in the “Open prices only” model on an M1 chart — with results more than 95 % identical to the “Every tick based on real ticks” model.

You should know: the “Open prices only” model on M1 always uses a knowingly understated spread (the minimum spread of each minute bar), so re-running a result on real ticks almost always shows a somewhat lower profit. We have done everything possible — and largely succeeded — to keep the overall test profile intact and matching between the two models in most cases.That is why we pay no attention to “real ticks” during optimization itself — and only at the very last stage, before taking a set live, we always re-verify it in the “Every tick based on real ticks” model specifically (not “Every tick”, which is the same M1 bars “from another angle”: ticks generated by the terminal's algorithm). The EA has a built-in mode for exactly this re-verification — §6.5.

Using the engine is as simple as it gets: set a pass budget in the Pass budget input, select the “Slow complete algorithm” mode with the Custom max criterion — and press Start (§5).

1.7The reliability — profitability axis

Sets used for trading live on a “reliability — profitability” axis. A set's position on this axis is determined by several factors, each on its own and in combination:

• The optimization period. Sets found on the last half-year of history are, on the whole, considerably more profitable than sets found on the last 2 years. Our premise is that the market changes constantly, and it is the most recent history that reflects the currently active tendencies; last year's tendencies may well be dead by now. This is no fringe theory — it is the well-established Adaptive Market Hypothesis (AMH). Hence a set will usually not work in the tester outside its optimization period (or will work with a deeper drawdown) — that is normal behaviour. A longer period shifts sets toward reliability at the cost of profitability; a shorter one — the other way. The golden mean is suggested by the number of trades: about 1,000 trades or more per test period looks good even for a relatively short period (half a year or a year), provided the planned trading time for the set does not exceed ¼ of the test period and the profitability of that quarter satisfies your risk-reward criteria.
• The forward check (the Forward input) — reserving part of the period for out-of-sample verification, §6.1.
• The drawdown ceiling (the Drawdown to % input) — the lower the allowed drawdown, the lower the overall profitability of the resulting sets and the higher their hypothetical survivability, §6.2.

1.8Broker and account-type binding

Settings produced by AI Pro100 optimization are usually specific to a particular quote source — that is, to a given broker and account type. We run our tests and publish sample settings for the RoboForex broker, account type Pro. When trading with another broker — or even another account type of the same broker (e.g. RoboForex ECN) — re-optimization for the quirks of that account's quotes may be required. This is a normal process: the robots exploit intraday market dependencies that can differ noticeably even in quotes of the same instrument from different sources. Obviously, the resulting SSNs should be traded with the very broker and account type on whose quotes they were obtained.

1.9The Rating quality score

The AI Pro100 engine uses its own “smart” quality score for a test pass — Rating; all optimization results are ranked by it. To see it in the tester's Optimization tab, select the Custom max criterion before starting (the drop-down next to the optimization mode on the Settings tab).

A positive Rating is earned only by passes that: are profitable over the whole test period; had no Stop out / Margin call events; and satisfy the Optimization group inputs set before the start — passed the forward check (if Forward is not 0 %), stayed within the Drawdown to % ceiling and collected at least Min deals trades.

The Rating formula intelligently combines: geometric-mean profitability; the number of trade series (unique entries); profit uniformity — equity curves made of a few large one-off “shots” get a low rating, other things being equal; and the maximum relative equity drawdown (a metric MetaTrader 5 does not show in the Optimization tab but reports on the Backtest tab as Equity Drawdown Relative).

2EA inputs

There are just ten external inputs, identical in both EAs — three groups in the properties window. Everything else is carried by the SSN set code (§1.5).

* The seeding itself is reproducible; the final optimization result is not guaranteed to be identical: what grows out of that “seed” depends on the order in which agent results arrive in the MetaTrader 5 tester, which cannot be predicted exactly.

3Quick start: the mql5.com demo

You can try the full AI-optimization cycle before buying. Per marketplace rules the Market demo does not execute optimization — so the EA itself hands you a dedicated tester build free of that limitation.

4Regular backtest

A single Strategy Tester run: checking a ready-made set on history — a sample set from the product page, or a result of your own optimization.

4.1The tester's Settings tab

4.2The EA's Inputs tab

• SSN code — paste the set code. Left empty (with Pass budget = -1) the built-in demonstration set is run.
• Minimal deposit — 0 (risk scale from the set code). A value >0 overrides the set's MM — a way to try the same set at a different risk scale.
• Equity SL % / Equity TP $ — optional: they work in the tester too, so you can rehearse your protection and target plan right away (§1.3–1.4).
• Everything else — defaults.

4.3The journal after the test

Progress and results are printed to the tester agent's journal (the Strategy Tester's Journal tab). The key lines:

The MK3 RC pass line is the quality summary: Rating (§1.9), annual geometric-mean profitability, maximum relative drawdown, number of trades. The full test statistics are in the tester's standard Backtest / Graph tabs.

5AI Pro100 optimization

The basic scenario: find fresh sets for your instrument, broker and period. Everything happens inside one Strategy Tester optimization — no external software.

5.1Preparation and settings

5.2Optimization progress: the Experts journal

Cycle messages go to the terminal's Experts journal (not the tester's). A normal run looks like this:

• The first line is the run's settings summary: budget, forward, MM ladder, filters.
• The building book, please wait... line appears when the passes are done and the EA is assembling the results workbook; on big budgets this takes minutes — the notice also hangs on the chart, keep the terminal open.
• The showcase is the 26 sets with the highest Rating: place | cluster (C<k>, §5.5) | SSN | Rating | Profit | drawdown | profit factor | trades | idx — the row number for reproduction.
• The last line is the full path to the xlsx workbook.

In the tester's Optimization tab the Result column is the Rating. Zeros in Result are normal: they are shown by passes that failed the filters or have no relevant rating (§6.2); the true values of all passes are in the book.

5.3The results workbook (xlsx)

The workbook is the file <Robot>_<symbol>_<datetime>.xlsx (e.g. Xaurus_XAUUSD_20260830120000.xlsx) in the MQL5\Files folder of the terminal data directory. The exact path is always printed in the journal. Workbooks accumulate — old ones are not deleted.

The book has four sheets, in this order: Weighted, ALL, ALLGM and Info. The first three are the same results table in a different selection and order; the fourth is the run's passport.

The Weighted sheet — start your selection here

Weighted is the first sheet of the book and, in our view, a balanced compromise between rating and profitability, based on the overall quality of the optimization results obtained.

What it does, in plain words: from every logic (cluster, §5.5) it takes the top-rated and the most profitable sets, ranks what it took by profitability (GMon %) and cuts off the rows that are weak by rating. Finally, sets with practically the same balance curve are collapsed — the most profitable one of the group remains.

Columns of the Weighted, ALL and ALLGM sheets

The ALL, ALLGM and Info sheets

The ALL sheet holds the rows by descending Rating: take it when rating matters most. The ALLGM sheet has the same columns but is ordered by descending GMon %: when profitability matters most. The contents of ALL and ALLGM differ — each sheet has its own selection: out of a group of similar sets ALL keeps the best by Rating, ALLGM keeps the most profitable one, and the 100-rows-per-cluster cap is counted by each sheet's own metric. So the number of rows may differ between the sheets, and a row of one sheet need not be present on the other.

The book holds only the rows that passed selection: a row that failed the Drawdown to % / Min deals filters (§6.2) or, in risk mode, a set with no working ladder level (§6.3) does not enter the book at all.

The book removes clones in two ways. Full duplicates (DD % and Profit equal to 2 decimal places) are collapsed at once. Besides them, the book collapses sets with practically the same balance curve — one set of such a group remains in the book: on the ALL sheet the best by Rating, on ALLGM and Weighted the most profitable one.

The book's size is capped: each cluster (§5.5) contributes at most 100 rows to each sheet — on ALL the 100 best by Rating, on ALLGM the 100 best by GMon %, and on Weighted the 50 best by Rating and the 50 best by GMon % (if one of the lists is shorter, the remainder is taken from the other). That way the book keeps both the top-rated and the most profitable sets of every logic, while the long tail of near-identical variations does not bloat the file.

The Info sheet is the run's passport: EA name and version, symbol, period, deposit, leverage, budget, search seed, filters and service counters; among them filtered_out — how many rows the filters rejected, cap_out_all and cap_out_gm — how many rows the 100-rows-per-cluster cap cut off on the ALL and ALLGM sheets.

5.4Reproducing a row

Any optimization result reproduces bit-for-bit as a single test. Three ways:

• Double-click the row in the Optimization tab — the terminal launches a single test of that row itself.
• The idx number from the book — type it into Pass budget of a single test (requires the service folder of the last run, which is kept automatically).
• The SSN code — copy it from the book or the showcase into the SSN code input. The everlasting way: the set is fully embedded in the code and depends on no run files.

At the end of every single run the journal prints Pass set: TF M5, cluster 7 — the signal timeframe and cluster of the set (the book has no TF column — the timeframe is optimized like any other parameter and lives inside the SSN).

5.5Clusters: 10 trading-logic groups

Every book and showcase row carries a cluster number — 1..10. A cluster is a stable group of trading logic: sets of one cluster trade in fundamentally the same way, sets of different clusters — in fundamentally different ways. The number is a pure function of the set itself: it depends neither on the run nor on the instrument, so books of different runs are comparable. What exactly stands behind a number is not disclosed, and the number does not imply quality: C3 is not “better” than C7.

The top-26 showcase is assembled round-robin over clusters (up to 4 places per cluster) — so the top holds candidates of different logic, not 26 variations of one set.

There are no per-cluster sheets in the book: the cluster of every row is in the cluster column of all three results sheets — Weighted, ALL and ALLGM; to look at a single cluster, filter or sort the book by that column in Excel. The 100-rows-per-cluster cap (§5.3) applies here too, and on every sheet: a large cluster contributes exactly 100 rows to a sheet.

5.6Picking a set from the book

• Start with the Weighted sheet — in our view it is a balanced compromise between rating and profitability, based on the overall quality of the optimization results obtained (§5.3).
• Then the ALL sheet (best Rating on top) if rating matters more to you, and ALLGM (best GMon % on top) if profitability does. Check the drawdown (DD %) against your risk tolerance and your Equity SL % plan (§1.3).
• Number of trades: aim for ~1,000+ per test period (§1.7).
• For a portfolio — different clusters (§5.5).
• Run the selected candidates through the precise mode — §6.5 — and only then allow them into live trading (§8).

6Advanced optimization

The tools that harden the result: the built-in forward, output filters, the risk ladder, pausing/resuming a run, and the final cross-check of the best sets under precise modelling.

6.1Forward testing: Forward

The Forward input reserves 25, 50 or 75 % of the test period for the classic out-of-sample check — without leaving the single optimization. Our difference from the classic approach: the search uses the latest history (closest to “now”) while the check uses the earlier part (§1.7: fresh patterns matter more than stale ones; the earlier history serves as an extra trial).

Example: period 2024.07.01–2026.07.01, Forward = 50%. Sets are searched on 2025.07.01–2026.07.01, then the best found are automatically re-run over the whole 2024.07.01–2026.07.01 — survivors are those that also made money on the earlier half.

• The pass axis widens automatically — forward passes are added on top of the budget, nothing to configure.
• In the Optimization tab the search (back) passes show Result = 0 while the forward passes show their forward Rating: sorting by Result gives a clean forward picture at once. The true ratings of back passes are not lost — they are in the book.
• The forward-run book: survivors by descending forward Rating (non-survivors do not enter the book); a collapsed Back column group (the “+” button) keeps each row's search-phase metrics. The sheets are the same four — Weighted, ALL, ALLGM, Info; the Weighted sheet of a forward book is built by the same rule (§5.3).
• In our own practice we mostly use 25 % or 50 %. Forward is off by default and not used in the sample sets we publish.

6.2Output filters: Drawdown to % and Min deals

Drawdown to % — the ceiling on the maximum relative drawdown: a pass with a deeper drawdown is deemed unfit. Default 100 — filtering off. The lower the threshold, the lower the overall profitability of the resulting sets and the higher their hypothetical survivability (§1.7).

Min deals — the minimum number of trades per pass (the Trades column): a result below the threshold is statistically empty and gets rejected. Default 100; 0 disables the threshold.

How the filters act: a failing row does not enter the book (the filtered_out counter on the Info sheet says how many), is left out of the journal showcase and shows 0 in the Result column of the Optimization tab. The search itself also respects the filters: no budget is wasted on branches that clearly cannot pass the thresholds.

6.3Risk ladder (Minimal deposit sweep)

The optimization checkbox on the Minimal deposit input with Start/Step/Stop values enables the risk phase: after the main cycle the best sets are automatically replayed on every MM level of the range, from the base downwards. The smaller the MM, the larger the lot relative to the deposit: the ladder down is a ladder of increasing risk.

• The base (top anchor) is the maximum of the range; it must not be below the Deposit of the Settings tab. Axis example: Start 10000, Step 1000, Stop 2000.
• For every set the EA picks the working level — the most aggressive MM at which the set still passes the §6.2 filters and does not blow up; it lands in the book's MM min column, and the row's SSN already carries that MM.
• The level history is a collapsed column group (a “G/D” pair per level): red fill — the level blew up, cut — a level below the break was not computed (pointless), an empty pair — no data.
• A candidate for which no level fits does not enter the book at all. Sets that never got a ladder (beyond the candidate-selection computation budget) do stay in the book: they rank in the common order with their base MM and an empty level history.
• In risk mode the journal showcase shows rows with a found working level — trade-ready sets straight from the journal.

The ladder can also be built on top of an already finished run: leaving every other input untouched, tick the MM checkbox with a range and press Start — the main cycle is not recomputed, only the ladder levels run live. The overlay's book is written next to the run's book (suffix _ovl1, _ovl2…); during an overlay the Optimization tab shows zeros — look at the book.

6.4Pausing and resuming a run

• Stop during an optimization is a pause: computing agents stop within seconds, progress is saved, no book is built. In the journal: AI Optimization INTERRUPTED at 1385 of 20000 passes: progress saved…
• To resume — press Start again with the same settings: the EA recognizes its interrupted run, restores the accumulated state and finishes the remainder (already computed passes come back instantly from the tester cache). In the journal: …resuming: 1625 of 20000 passes restored, continuing. Resume as many times as you like.
• Any change of the inputs that shape the run (budget, Magic, Forward, filters, ladder, Timing profile, Equity SL/TP, symbol/timeframe/dates) makes it a different run. If that would kill the interrupted run, the first Start refuses with advice to restore the previous settings; a second Start with the changes kept is a conscious fresh run.
• Pressing Start again after a normally finished run with the same settings is refused (nothing to resume, the tester cache would replay everything at once): clear the cache (right-click the run in the Optimization tab → delete) or change Magic; if the cache is already cleared — just press Start a second time. Tip: when making several similar runs in a row, bump Pass budget or Magic (seed) by one each time — then the cache never needs any attention.

6.5Precise-mode confirmation: Pass budget = −N

The final step before live trading (§1.6): re-run the best book rows under “Every tick based on real ticks” and get marks in the book — who got confirmed.

The outcome is the Pass column (the last column of all three results sheets — Weighted, ALL and ALLGM): a green ✓ — the set is confirmed by the precise model; a red ✗ — not confirmed; empty — the row was not checked yet. Repeated -N launches accumulate the marks: failed rows are not taken into the following checks, so -20 and then another -20 test the next portion of candidates. Successive launches form a selection funnel.

6.6Risk sweep of one set: SSN × Minimal deposit

How does one ready-made set behave across risk scales? Paste its code into SSN code, tick the optimization checkbox on Minimal deposit with a Start/Step/Stop range — and start the optimization. The AI layer stays out (a single journal line, AIO passthrough: plain SSN x MM sweep…): it is a plain MT5 table where every pass is the same set with an MM from the axis, and the Result column is its Rating at that risk level. Modelling and period — any; a sweep has no book and no showcase.

7Server time profile (Timing profile)

AI Pro100 sets exploit intraday market patterns — and the market's “internal day” does not match the broker server's clock, drifting against it twice a year with daylight-saving shifts. The Timing profile input tells the EA your server's time rule so the set's trading window always points at the same market phase.

7.1Why it matters

Forex and metals live on the New York schedule: the trading day ends at 17:00 New York time. Most brokers' servers, meanwhile, run on Eastern European time with the EU daylight-saving calendar. Europe and the US switch their clocks on different dates — so for 3–4 weeks a year (2–3 weeks in spring, a week in autumn) every market session is shifted by an hour in server clocks: about 7 % of trading days. With the right profile the EA recomputes its trading window onto “market days”, and the window rides through those weeks without shifting.

The market-day anchor is determined automatically per instrument and is not configurable: cryptocurrencies count from 00:00 UTC, everything else from 17:00 New York. The applied profile and anchor are printed on every start:

7.2Profile values

7.3Determining your broker's profile

• The most reliable way — ask the broker's support: “what time zone does the trading server use, and does it switch clocks (by the EU or the US calendar)?”
• A practical clue: compare the server time (the Market Watch window header) with UTC. A difference of +2 in winter and +3 in summer — EU DST +2/+3. A constant difference all year — one of the Fixed values. Daily bars closing exactly at the New York close both in winter and in summer — US DST.
• In doubt — Raw quotes timing: the behaviour of previous generations, without calendar arithmetic.

7.4Rules of application

• Optimization only. The input takes effect when an AI cycle starts: the chosen profile is written into every produced set. Choose it before the start — a finished code's profile cannot be changed later (only by a new optimization).
• Testing and trading. The profile comes from the SSN itself and the input is ignored — the EA merely reports in the journal what it applied.
• Sets of previous generations (codes without the profile field) work as before — on raw server time.

8Live deployment

Taking a verified set to a real account: preparation, monitoring, and what to do in emergencies.

8.1Account preparation

• Broker and account type — strictly the ones whose quotes the set was optimized on (§1.8).
• Leverage — the maximum available from the broker on this account type. AI Pro100 is designed to use leverage to the fullest: the higher the leverage, the farther the Margin call / Stop out levels — and the lower the chance of hitting them. The set's survivability depends on this directly.
• A separate account, for the EA only. The Equity SL/TP mechanisms look at the equity of the whole account: parallel manual trading or other EAs on the same account distort the protection, the target and the trading itself.
• Deposit — not below the set's MM (the MM min column of the book / the mm value in the Set applied log line). Risk scales from the MM: the same set on a larger deposit trades a proportionally larger lot.
• Continuity. The terminal must run uninterrupted — a VPS or a stable PC with reliable internet. Algo Trading enabled in the terminal.

8.2Launch

8.3What you see while it works

• Virtual levels on the chart — the EA's working SL/TP (§1.2). Positions in the terminal carry no stop-losses — that is normal.
• The exception is an active Equity SL %: a common broker-side stop-loss appears on all the EA's positions, recomputed on every new opening. Don't be surprised by it and don't touch it.
• The EA panel on the chart. The Pause button suspends the trading logic (the EA opens and manages nothing); pressing it again resumes. The other panel elements are for special cases — leave them alone in normal operation.

8.4Restarts and recovery

A terminal or VPS restart is not a problem: on start the EA restores the state of the current series (from its state file and the account's open positions) and carries on. If a mismatch is detected during recovery (e.g. positions changed while the EA was off), it reports this with an Alert and a journal line — figure out the cause before continuing.

8.5Emergencies: the EA has stopped

The Robot stopped functioning. notice on the chart means the EA has halted: positions closed, trading stopped. Possible causes and actions:

Resuming after a halt is a deliberate act: re-attach the EA to the chart (or change its inputs — that reinitializes it) once the cause is understood and the decision is made.

8.6Scheduled set rotation

The AI Pro100 working cycle (§1.4): trade a set for roughly ¼ of its test period or until the target is reached — then re-optimize for the latest market and start again. Do not get attached to a spent set: the system's strength is cheap and fast self-optimization, not “eternal” settings.

9Appendix: messages and common questions

AI Pro100 by Xaurus.Pro © 2025–2026 · Trader's Guide v1.2 (September 2026). Expert Advisors: Xaurus AI Pro100, Swinger AI Pro100. Leveraged trading carries high risk: use only funds you can afford to lose (§1.1).