TradeWarden v1.20 — test report

Built by Alireza.

Generated 2026-10-07 18:29 (VPS time). Every row in the tester tables is parsed from a real Strategy Tester run of the final build; nothing is estimated. The live section lists what was measured on MetaQuotes demo accounts with real order execution.

Tested build: TradeWarden.ex5 1.20, SHA-256 57ac74f0ffd739746ca5c08b86eb92ba7dc7d7e44d130a3efabe513f96108497. The file in the package has the same hash.

Summary

Account: $10,000 and USD unless the row says otherwise, leverage 1:100, MetaQuotes-Demo history. A check is one assertion evaluated by the self-test (for the management scenarios it runs on every tick). An event is a real action of the EA that the test saw (a stop moved, a partial closed, a lock fired, a strike counted).

Runs that did not pass: 4. 2 failed:

• Hedging · 1-minute OHLC · 2026-08-10 → 08-15, S57 EURUSD (FAIL): The 1-minute OHLC model jumped over the keep line in one step. A 0.5-lot buy opened at 1.15428 on 2026.08.11 00:00; the M1 bar of 00:01 has a 7.4-pip range (high 1.15490, low 1.15416, from the server's history), and the model goes from the bar's high to its low with no tick between: the day's best was +31.00 at the high and the next tick was -6.00 at the low. The guard locked on that first tick below the keep line (15.50), as it should; the test allows 15.00 below the line and the step was 21.50. S57 passed in the September week (hedging and netting) and with real ticks. This is the limit 'Protect today's profit acts on ticks'.
• Netting · 1-minute OHLC · 2026-08-10 → 08-15, S57 EURUSD (FAIL): Same as the hedging run of this week: the same 0.5-lot buy at 1.15428 on 2026.08.11 00:00, and the 1-minute OHLC model went from the M1 bar 00:01's high (1.15490, the day at +31.00) to its low (1.15416, -6.00) in one step. The guard locked on that first tick below the keep line (15.50), as it should; the test allows 15.00 below the line. S57 passed in the September week on both accounts and with real ticks.
• Hedging · 1-minute OHLC · 2026-07-06 → 07-11, S23 EURUSD (NO-CHECKS): The test's own trades (58 reversing 2-lot trades) lost 2,285.40 in that week (balance 10,000.00 -> 7,714.60, the lowest 7,060.60), so equity never reached the 0.3 % target (10,030.00): the target lock had nothing to do and no check ran. In the September and August weeks, on both accounts, S23 locked exactly at the target (equity 10,030.00). Same as in 1.10.
• More symbols and an EUR account · hedging · 1-minute OHLC · 2026-09-14 → 09-19, S1 XPTUSD (NO-CHECKS): S1 waits for a spread of at most half its smallest stop (5 pips: 2.5 pips) before its first trade. On MetaQuotes-Demo platinum's smallest spread in that week was 617 points = 61.7 pips (median 2,857 points), read from the same 6,598 M1 bars the tester generated, so S1 never traded and no check ran. Same as in 1.10.

Mutation tests: does each test catch its bug?

Each row switches one fix or rule off in the source, builds, and runs the scenario that guards it. The scenario must FAIL; then the source is restored (byte for byte) and rebuilt. A test that still passed with its rule switched off would prove nothing.

57 of 57 mutations caught, source restored after each.

What each scenario proves

Every run

Hedging · 1-minute OHLC · 2026-09-14 → 09-19

Netting · 1-minute OHLC · 2026-09-14 → 09-19

Hedging · real ticks · 2026-09-15 → 09-17

Netting · real ticks · 2026-09-15 → 09-17

Hedging · 1-minute OHLC · 2026-08-10 → 08-15

Netting · 1-minute OHLC · 2026-08-10 → 08-15

Hedging · 1-minute OHLC · 2026-07-06 → 07-11

Other guard settings · hedging + netting · 1-minute OHLC · 2026-09-14 → 09-19

More symbols and an EUR account · hedging · 1-minute OHLC · 2026-09-14 → 09-19

Final build (the tested and shipped file), 2026-10-07

Every live test below ran on the final build (57ac74f0…, the same file the Strategy Tester runs above used), on the same hedging and netting demo accounts, with real orders. Times of fills and closes are the terminal's own trade journal (Logs\20261007.log), not the EA's.

Every MetaTrader shutdown in this run was clean: the Experts log has no "Abnormal termination" from the start of the run to its end (the earlier runs that day had 9, see Limits).

Bugs the tests found in 1.20 (all fixed in this build)

Limits

• MetaTrader cannot refuse a manual order before it is sent. The panel refuses before sending (no order reaches the server: measured during the wait after a loss). A trade from the MT5 phone app, the web terminal or one-click trading is closed right after it opens, so its spread is lost. This is true of every MT5 guard.
• Removing TradeWarden from the chart, or closing MetaTrader, turns the guard off. The accountability partner (if set) gets a Telegram message when TradeWarden is removed; nothing can stop a trader from doing it.
• Telegram needs one step in MetaTrader: Tools → Options → Expert Advisors → Allow WebRequest for listed URL → https://api.telegram.org. No program can set it (no setting file or message does it). Until it is set, the panel and the log say exactly that (measured live: "Telegram is blocked by the terminal: add https://api.telegram.org …"). Phone push through the MT5 app needs only the MetaQuotes ID.
• The guard acts on ticks. Measured with real ticks (S5): a 5-lot test position on a $10,000 account lost $40 between two ticks. The lock fired on that very tick, but $5 under the floor, because the safety buffer ($10) was smaller than one tick's move. Set the buffer larger than what one tick can move with your largest position.
• Hidden stops fill at the terminal's latency (137 ms here). In a fast move the server's emergency stop can fill first (measured on gold in 1.10: $165 lost instead of $99).
• OCO is client-side. One quote can fill both sides; the later fill is closed at once.
• The news filter works on a live terminal only (the MT5 calendar is not in the Strategy Tester).
• The guard works only while MetaTrader 5 runs with Algo Trading on: a VPS for funded accounts.
• "Reset every lock" is copied into templates: never save a template while it is on.
• Netting, an add at a partial-close level: volume added in the same instant the price crosses a partial-close level can make that one partial close use the old volume. The window is one cycle (under 0.5 s); not reproduced in a test.
• Automated clicks: in the live UI runs, a few posted clicks (about 1 in 40, always seconds after another click that re-drew the panel) never reached TradeWarden (the button showed pressed, nothing else happened); clicking again worked. The same clicks with 3-second gaps all worked (8 of 8 tab changes at 100, 115 and 130 %). It looks like a property of posted window messages, but it was not proven with a physical mouse.
• "Abnormal termination" while MetaTrader closes: in the test runs before the final one, MetaTrader wrote it for TradeWarden 8 times on 2026-10-07 (and once for a test script), always while a test script was closing the terminal: TradeWarden was still busy when MetaTrader stopped it, and its exit lines were missing. In the final live run: none. The cause was not found; a suspect is the UI tests' hook, which asks the chart about every object while the chart is being closed. The state file is written on every change, so locks and strikes are kept (the restart test shows it); only the "MetaTrader was closed" alert to a partner may be lost in that case.
• A break cannot be cancelled from the panel, but it is not a vault. Removing TradeWarden from the chart ends it, and so does the Reset every lock input; both send the accountability partner a message (if one is set).
• Protect today's profit acts on ticks, like every limit: a fast move between two ticks can take the day below its line before the lock fires; with close every trade on, the trades close at market.
• Trade only these symbols matches by prefix, so a symbol's suffix (EURUSD.r) is covered; list full names (EURUSD, not EUR).
• FundedNext's 3 % rule: FundedNext defines the risk as "the maximum potential loss on a trade at a time, based on stop-loss placement, the maximum realized and unrealized (Floating) losses across all trades, calculated against the initial account balance". TradeWarden caps the stop risk of the open trades at 3 % of the initial balance. If FundedNext also counts losses already realized, keep a margin. A trade without a stop has unknown risk, so with this rule the panel refuses new trades until every open trade has a stop.
• The discipline score counts what TradeWarden enforces: a Live Guard strike, or a manual or panel trade your rules closed or cut. A rule you keep only in your head is not in it.
• Monitor only does nothing: every rule alerts once a day per rule and trade, and nothing is closed, cut, refused or locked. The checklist on the panel still asks.
• Phone notifications hold 255 characters (MQL5 documentation): a longer daily summary or weekly report is sent in parts of whole sentences. Not tested with a real phone.
• Hidden pending orders live in TradeWarden, not at the broker: while MetaTrader is closed or not connected, a hidden order is not sent when the price reaches it. Use real pending orders when you cannot keep MetaTrader running.
• At a terminal start no rule is checked until the account's data is there (OFFLINE on the panel), usually a few seconds after the connection; a trade opened in those seconds is judged when the guard starts.
• This is version 1.20 with no users yet.

Not covered by tests yet

These work in the code but no test exercised them; treat them as untested: - Telegram to a real chat: the message format was checked against Telegram's API with a fake chat ("chat not found"), and the blocked-address hint live; no message was delivered to a real chat in these tests. MT5 phone push: no MetaQuotes ID is set on the test terminal. - Windows display scaling of 125 % / 150 %: only the panel's own size buttons (90–175 %) were tested; the code scales by the screen's DPI. - A trial that expires while it runs: the date is checked when TradeWarden starts, so a running guard is never switched off in the middle of a trade; it stops at its next start. Only "expired at start" was tested live. - Position sizing settings: risk as "% of equity", and a non-zero commission per lot. - News filter: the 30-second retry when the calendar is not loaded yet. - Pending-order trailing skipping straddle orders. - Close-all retry with a real closed market (tested with refusals simulated in the Strategy Tester, S41); OCO pairs with a broker that really rewrites comments (tested in the Strategy Tester, S40). - Symbols and currencies: indices (MetaQuotes-Demo has them with trading disabled), crypto (not offered there), platinum (see the runs that did not pass, if any), and account currencies other than USD and EUR. - Failover by heartbeat: a chart that stops responding without closing (the 15-second path). - Third-round rules on a live account: trading hours, days and direction, the spacing and trades per hour, the max lot, the per-symbol loss cap, the time stop, fixed lots, the ATR stop, pending expiry, the journal, the weekly report and the discipline score were tested in the Strategy Tester (S60–S68, hedging and netting). Live, on the demo account: the checklist with real clicks, the GUARD rows, monitor-only mode, a dragged planning line, a start without a connection, the chart vault (another EA's chart closed while locked and opened again, with its EA) and the light theme (screenshots). Stop-limit, hidden and grid orders and the lock password: in the Strategy Tester only (S69–S72, hedging and netting; the password on hedging). - Second-round rules on a live account: a break for the rest of the day across a real daily reset, protect today's profit, the symbol list and FundedNext's 3 % rule were tested in the Strategy Tester (S56–S59, hedging and netting, three market weeks and real ticks); live, only the 15-minute break was taken (with real clicks).

Against the market (product pages read 2026-10-05 to 2026-10-07)

The side-by-side table of 16 tools is in COMPARISON.md: 58 rows of what a trade manager and a prop-firm guard do (the 51 rival features TradeWarden has and 7 only TradeWarden has), then the 3 features of the table it does not have and, beyond the table, what Trade Assistant 38 in 1's page lists that TradeWarden does not have (one-click copy or reverse, close by, price alerts, scheduled orders, chart tools) and the finer steps on Forex Trade Manager's page (break-even and trailing in up to 4 levels, limit pullback orders, splitting a large trade). A summary:

From each product's own page; a "–" means the page does not list it. Meridian is for NinjaTrader, not MetaTrader; its row is from its announcement post, which names a 0–100 score from 7 behaviour signs without listing them one by one.

Second round of research (2026-10-07). The guard with the most reviews on MQL5 is DrawDown Limiter ($99, 25 reviews, 5.0), whose panel adapts between wide, standard, dense and minimal layouts. Among the 70 newest MT5 utilities, about 25 were prop or risk guards, none with a review yet. We read 687 reviews of 18 MT5 trade managers and guards (every low-star review and the newest ones): in the 518 good reviews, support was praised most (154), then a tool that is easy or simple (96); in the 61 bad ones, a panel that freezes or lags MetaTrader (14) and a seller with no answer or no manual (14) were the top complaints, and one guard "didn't work with the limit I have set. I lost a 540 euros FTMO test". On TopstepX (futures), "More Trader Configurable Risk Tools" (daily limits, lockouts, a trade clock, a trailing daily limit, symbol blocks) had 641 votes on its feature board; MT5 has none of these built in.