Scalping Robot Pro 3.0 against 4.0: what a version bump actually changed

2026-09-03, based on our two dated catalog runs of v3.0 and v4.0 on XAUUSD M1 from 2003.05.05, both rated 100% real ticks. The audit page shows the verdict lamps for free.

The vendor published version 4.0 on 2026-08-14 with a one line changelog, Dynamic Lot Increase. Version 3.0 had arrived on 2026-07-07 with three claimed additions, an extra volatility filter, a news filter extended to FED events and a bank holiday filter. Our catalog holds a full tester run of each version on the same gold history, so the question of what an update did is not a matter of reading the release note. It is a matter of comparing two audits.

The answer is that almost nothing moved. Both runs come out with the same six verdict lamps, data quality ok, structure red, costs ok, concentration ok, regime caution and prop fit red, and with the same 4 red flags. Both print a head profit factor of 0.88, an expected payoff of -0.10 USD per trade, an end of day Sharpe of -2.58 and a break even shock of -0.18 USD per trade. After our commission retrofit version 3.0 nets -53,263.12 USD and version 4.0 nets -53,234.68 USD. That is a gap of 28.44 USD across 302,139 and 302,798 trades.

And even that gap is not the version. The 3.0 run ends on 2026.08.17 and the 4.0 run on 2026.08.19, because our protocol runs every window to the date of the run itself. Two extra days carry 659 extra trades, and our own comparison file records the limitation in plain words, the windows differ in end date so absolute totals are not directly comparable. What stays comparable is everything shaped as a rate rather than a total, and every one of those is unchanged.

What the input records show that the changelog does not

Our reports keep the full input block of every run. The catalog protocol fixes the test setup and the lot line, a 0.01 base lot on a 100,000 USD frame at 1:5000, and leaves the strategy inputs at what the build itself brings. Three inputs appear in the 4.0 record that do not exist in the 3.0 record, DynamicLotIncrease set to false, LotIncrease at 0.01 and StartIncreaseFromOrder at 5. The advertised feature of the release therefore ships switched off, which is the reason it cannot show up in any number below.

Two switches did move. EnableCandleSizeFilter is false in the 3.0 record and true in the 4.0 record, and EnableVolatilityFilter does the same. The trailing stop distance goes from 30 to 50 and the trailing step from 10 to 20. One thing we deliberately do not claim here. The 3.0 changelog announced an additional volatility filter, and the input named EnableVolatilityFilter sits at false in the 3.0 record and at true in the 4.0 record. We cannot map a release note line to an input name with certainty, so both facts stand as they are and the connection stays open.

Two filters on, and the basket behaves the same

If a candle size filter and a volatility filter change what a grid does, the structure forensics is where it would show. It does not show. Version 3.0 added 42,421 entries against an open losing basket versus 32,641 with the move, version 4.0 added 42,534 against 32,744. Both runs reach at most 5 simultaneous positions. Both hold long and short at the same time, 12,444 moments in 3.0 and 12,473 in 4.0. Both are flagged red for averaging into adverse moves and red for hedging, and both carry the regime note of 23 negative years.

The cost ladder is equally flat. Gross before costs the 3.0 run loses 29,111.71 USD and the 4.0 run 29,030.55 USD. Our retrofit charges 3.50 USD per lot per side across 604,278 and 605,596 deals, which is -24,171.12 USD and -24,223.84 USD. In both versions the commission line is about as large as the trading loss itself, and in both the balance check reconstructs the report to 0.00 USD, so the ladder is not an estimate.

What a 1,000 USD account does with either build

The vendor recommended deposit is 1,000 USD. Our survival map starts a fresh account of that size at every deal boundary of the run and replays what follows. In version 3.0 96.9% of 604,278 start points end in account death, in version 4.0 it is 97.0% of 605,596. Chained end to end, both histories burn 53 accounts before they reach the present.

Both audits also record exactly 1 drawdown episode, and it is the whole run. It opens on 2003-05-19 at a balance of 99,999.28 USD and has still not recovered when the data ends, 6,062 trading days later in 3.0 and 6,063 in 4.0, at 46,736.88 USD and 46,765.32 USD. Even the worst single trade is the same in both versions, 2011-12-13 at -97.92 USD.

Why we run the older build at all

A version number is a claim about change, and a catalog that keeps only the newest run cannot check it. Here the check is cheap and the result is specific. Between these two builds the measured behaviour of the strategy did not move, while the price our watcher recorded for the product did, 1,799 USD on 2026-08-15, 1,599 USD on 2026-08-19, 1,299 USD on 2026-08-22 and 1,599 USD again on 2026-08-26.

The honest limits. This compares two strategy tester runs in our documented data world, in sample by definition, and it says nothing about builds we have not measured. The product page listed version 5.0 on 2026-08-24 with the changelog Volatility Spike Detector and version 6.0 on 2026-08-25 with the changelog Fixed minor issues. Neither has an audited run published in this catalog yet, so neither appears in any number above. A backtest cannot prove an edge and it cannot disprove one either. What two backtests of two builds can do is show whether an update changed the thing it was announced to change.

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