Scalping Robot Pro Test: 82.9% to 12.6% Win Rate by Feed

Scalping Robot Pro Test Review — The Same Logic Goes from 82.9% to 12.6% Win Rate Depending on the Feed

We ran our own measurement-based test of Scalping Robot Pro (XAUUSD), a scalping EA sold on the MQL5 Market. The primary data in this article consists of measured results from running the original binary itself in the Strategy Tester. To analyze the internal logic, we additionally used a verification replica (a clone) that we built from behavioral analysis. The tables clearly state which figures come from which source. No official images or official performance records are reproduced here.

Test Conditions (Full Disclosure)

Result 1: Simply Changing the Feed Makes Performance Collapse Monotonically

These are the results of the same verification replica, with no settings changed at all and only the feed swapped.

The period is 2026 (for Exness and HFM, only the range for which data was available). The reports for this comparison do not record DD, so the max DD shown is the lower bound implied by the final loss percentage.

And when the original binary is run on the MetaQuotes demo, it produces an 82.9% win rate, PF 4.16, and +453.7% (this site's own measurement of the original binary: MetaQuotes-Demo, XAUUSD M1, January 2018 to June 2026, initial deposit $10,000, fixed 0.01 lots, total net profit +$45,365.76, max balance DD over the same period 7.56%). This figure is not a value published by the seller. In other words, the kind of strong performance you see on the sales page held up only on MetaQuotes demo history, and it deteriorated monotonically as tick quality approached real market conditions. For the general background on feeds and backtest models, see Real Ticks vs OHLC.

Result 2: A Direct Test of Whether the Entries Themselves Carry Information

Logic analysis (based on the verification replica) showed that this EA is structured as a mean-reversion counter-trend strategy (fading deviations of roughly 1.7–1.9 ATR) + a thin fixed take-profit of about $2 (86% of winning trades fall in the $1.9–$2.1 profit band) + a median holding time of about 10 seconds. We therefore directly measured whether the entry points themselves have an edge.

• Applying the original binary's set of entries to MetaQuotes data from a different date range lowered the win rate to 64–70%.
• On 6.87 million independent real ticks, all 24 grid-based decision rules produced win rates of 27.6–35.2%, nearly matching the random-walk theoretical value (33.3%).
• Running the original binary itself on real ticks gave a PF of 0.91 (measured) (period: one week, January 5–10, 2026; max equity DD over the same period 2.02%).

That is, within the scope of this site's measurements, no statistical edge could be detected from the entry signals on a real feed. As a textbook example of the "wins in backtests, collapses in live trading" structure, it corresponds to a real-world case of Why an EA Wins in Backtests but Loses Live.

Summary (Facts Only)

• The original binary's strong results (82.9% win rate / PF 4.16) were confirmed by measurement on MetaQuotes demo history (January 2018 to June 2026, max balance DD 7.56%).
• The same logic degraded monotonically as feed quality improved; on Dukascopy real ticks it showed a 12.6% win rate / PF 0.25 (P/L −99.9%, with max DD at or above that).
• The hit rate of the entries was equivalent to random on independent real ticks.
• When testing thin-take-profit scalping EAs, we strongly recommend retesting on real ticks rather than Model=1 or synthetic ticks.

All of this site's own EAs are tested on real ticks and multiple broker feeds, and the figures are published. See the Testing Methodology for our approach and the EA Ranking for the full lineup.

Notes on Naming and Rights

• Scalping Robot Pro and related names are trademarks or product names of their respective owners. This article uses them descriptively for comparison and critique, and has no affiliation with the seller or MQL5.
• Figures labeled "verification replica" are measured values from replica code created by this site for analysis purposes, not values claimed for the original product. Measured values from the original binary are explicitly labeled as such.
• If you find any factual error in this article, please let us know via Contact. We will verify and correct it promptly. The original data has been preserved.