Same program, three profit factors: the signal identifiers, the recomputation, and three corrections

When I published the launch post for Execution Caliper on 21 September 2026, its central example was one EA run on three accounts, showing profit factors of 1.08, 1.79 and 2.39. I promised a separate post with the signal identifiers, the recomputation and the three corrections to my own earlier tables, so that the claim could be checked rather than trusted. This is that post. When I did the recomputation, from the exported trade lists on my own disk, it contradicted parts of what I had written. So this post does two things. It gives the numbers and the method, and it withdraws the statements that the data does not support. Every figure below was recomputed on 21 September 2026 from those exports; where I could not recompute a figure, I say so rather than repeat it. The post contains no performance figures for my own products and no opinion on the EA in question. It is a record of what the data says and of where I was wrong.

Two terms are used throughout. Profit factor is the sum of the profits on winning positions divided by the sum of the losses on losing positions, so a profit factor of 2.0 means the winners earned twice what the losers cost. The execution tax, the ratio this tool measures, is the round-trip cost of a trade divided by the stop width of the strategy that placed it.

What I am withdrawing

Five statements in the launch post and in the product description are withdrawn. The first is that the three accounts covered the same period; two of them started eight and a half months apart. The second is that the only difference between them was the price at which orders were filled; matched trade for trade, the fill difference was too small to explain anything, and the win rates, which I said were the same on all three, can be checked for only two. The third is the figure 1.08 itself, which has no trade list behind it. The fourth is the phrase published statistics, used for tables that had never been published. The fifth, from the product description, is that most subscriber complaints about that product were about slippage; the public review page does not support it. Each of these is set out below with the numbers that replaced it.

The signals

Two of the three accounts are public MQL5 signals. Signal 2195619, "The Gold Reaper Live default settings", runs on PepperstoneUK-Live and started on 3 February 2024. Signal 2265877, "Gold Reaper New V2 2", runs on CapitalPointTrading-MT5-4 and started on 22 October 2024. Both exports contain XAUUSD positions only. I took the closed-position exports of both in the second week of August 2026: 1,283 positions for 2195619, from 7 February 2024 to 7 August 2026, and 830 for 2265877, from 29 October 2024 to 7 August 2026. Recomputed from those exports with commission and swap included, the profit factors are 1.794 and 2.400, against 1.79 and 2.39 displayed on the signal pages at the time; on the Profit column alone they are 1.814 and 2.457. The net win rates are 71.9 and 71.7 percent, the gross win rates 73.1 and 73.7. One caveat on the first account: its export carries a commission entry on only a small share of its rows, so its net figure may be slightly generous, and the value displayed on the signal page is the reference. Both accounts have kept trading since the exports were taken. On 22 September 2026 the two pages showed 1.81 and 2.34, with 1,323 and 870 trades and win rates of 72.33 and 72.41 percent.

The third figure, 1.08, is different in kind. It comes from a Myfxbook account for the same program, quoted on a third-party website. I have no trade export for it, the signal pages do not show it, and I could not verify its period, its broker or its settings. It should not have stood beside the other two, and it is not used anywhere below.

What the exports showed

The launch post said the three accounts covered the same period and differed only in fill prices. The exports do not support that. Signal 2195619 started eight and a half months before 2265877, and its first 424 positions, from February to October 2024, had a gross profit factor of 0.924 on their own. Cut both exports to the window they share, 29 October 2024 to 7 August 2026, and the net profit factors are 2.396 for 2195619 and 2.400 for 2265877, with net win rates of 72.4 and 71.7 percent. The gap between 1.79 and 2.39 was the start date, not the account.

To check whether the two accounts took the same trades, I matched positions one to one by direction and opening time within three minutes: 760 pairs, 89 percent of one export and 92 percent of the other. Across those pairs the entries on 2265877 were less favourable by a median of four cents per ounce, mean eight cents. For the 425 pairs that also closed within three minutes of each other, the round-trip difference had a median of zero. With an average losing trade of about $12.60 on 0.01 lot, the execution gap between these two accounts was under one percent of a typical loss. The matched pairs give net profit factors of 2.355 and 2.348. So the fill prices did differ, but by an amount that changes nothing in this example, and I had pooled two periods in a paragraph that warned against pooling periods. The launch post also said the win rate was the same on all three accounts. Only two can be checked; those two are 71.9 and 71.7 percent net, and the third account cannot be checked at all.

The review count

The product description said that most subscriber complaints about that product were about slippage rather than the strategy. On 22 September 2026 the review page of signal 2265877 showed ten reviews. Seven describe trades that were not copied to the subscriber's account at all, one describes a symbol-mapping failure, and two mention slippage. Trades that do not copy are an execution problem of a different kind, and not one this tool measures. The sentence is withdrawn. A count like this also has no place in a permanent product description, because it changes with every new review; it belongs here, with its date.

Correction one: mixing symbols in one distribution

My first reference table, a working table behind the first design, took nineteen public signals with 69,236 closed positions and computed one execution tax per signal. The stop width was taken as the 90th percentile of losing-trade size in pips, that is, the loss that only one losing trade in ten exceeds, across every symbol the signal traded. A pip of EURUSD, a pip of XAUUSD and a point of an index are not the same unit, so in a multi-symbol signal the symbol with the largest numbers dominated the percentile, and the ratio came out small. Recomputed per signal and per symbol, with at least thirty losing trades per cell, there are 91 cells with a median of 2.40 percent, a 95th percentile of 13.06 and a maximum of 19.99. Inside one signal, 2361862 with nineteen scored symbols, the ratio runs from 1.74 percent on XAUUSD to 19.99 on NZDCHF, a factor of 11.5, and on 2373305 from 0.11 to 5.16, a factor of 46.9. The same EA on the same account can carry a small execution tax on one symbol and a large one on another. Only the single-symbol rows of the first table survive unchanged; the multi-symbol rows cannot be regenerated from any script I kept, so the table is withdrawn rather than patched. That table was never posted on this blog, which is why the phrase published statistics in the launch post and in the product description was wrong. The tables were mine and they were earlier, but they were not published.

Correction two: thresholds chosen after the data

The first design had a green, amber, red verdict. The green boundary was set at eight percent because the largest per-signal value in that first table sat just below eight; the line was drawn from the data it was meant to judge. I had pre-registered a rule to drop the verdict if the recomputed boundary moved above fifteen percent, but I had not registered how many losing trades a cell needed before it counted. With thirty, the maximum is 19.99 percent and the verdict dies; with one hundred, 17.54 and it dies; with two hundred, 13.44 and it survives. A rule whose outcome depends on a choice made afterwards is not a rule, so the verdict was removed and the tool reports the ratio and no verdict.

Correction three: pooling periods and brokers

Signal 2358336 trades mainly EURAUD and AUDCAD with fixed stops. Across all seven and a half years of its history its EURAUD ratio is 4.00 percent; excluding 2020 it is 14.12. In 2020 the median and the 90th percentile of its losses were both exactly 50.0 pips, and the stop-loss distances in the export step from about 37 to 50 pips in the early years to 101 pips from 2024 on, so one ratio over the whole history described neither generation of settings. On 2361862, the NZDCHF cell that produced the 19.99 percent maximum has 89 of its 112 positions in 2026 alone.

My assumption about brokers was further from the data. The tables used one assumed round-trip cost per asset class for every broker, 1.0 pip for a major pair. Spreads can be measured from an export: a buy opens at the ask and closes at the bid, a sell the opposite, so where one symbol shows an ask-side and a bid-side price in the same second, their difference is the spread. On signal 2384807, which pays no commission, EURUSD measured 3.25 pips, NZDUSD 3.50 and USDCHF 3.45, and its EURUSD ratio moves from 1.05 to 3.43 percent. On 2379915, which pays commission separately, AUDCAD measured 0.85 pips against an assumed 2.0. For crypto a flat 15 dollars is 1.65 basis points of one BTC contract and 56 basis points of one ETH contract; measured, the ETH spread on 2273131 is $1.94, and its ratio falls from 13.44 to 1.74 percent. These same-second pairs are few, between four and seven per symbol, so read them as orders of magnitude, not as precise values. The direction of the error is the point: an assumed cost can be several times too low or several times too high, and the ratio moves with it.

What changed in the tool

Each of the three errors now has a rule inside the instrument. Stop width is taken per symbol, never as a pooled percentile. Costs are measured on your own account and nowhere else; there is no reference table. When the data is missing, the panel prints not measured instead of an estimate. The three-colour verdict is gone. A red line at about twenty percent of the stop width remains on the cost bar, because that is where my own EAs died of execution cost in live measurement; it is a fact about my EAs and says nothing about anyone else's.

What has been corrected where

As of 22 September 2026, the paragraph headed Why it exists in the launch post has been rewritten and carries a note recording what it said before, and the product description on the Market has been rewritten in the same terms, with earlier working tables in place of published statistics. The launch post's own sentence about published statistics is being changed to the same wording. Where an older copy of either text disagrees with this post, this post is the record. This post was itself revised on 22 September 2026, to add the section on what is withdrawn and to correct two figures, the review count and the number of same-second pairs.

How to reproduce

The trade lists are the positions exports of the two signals, at /en/signals/2195619/export/positions and /en/signals/2265877/export/positions, which need a logged-in session; mine were taken in the second week of August 2026, and the last position in each is dated 7 August 2026. Profit factor is the sum of profit over winning positions divided by the absolute sum over losing positions; net includes commission and swap, gross uses the Profit column alone. The execution tax is round-trip cost divided by stop width, where stop width is the 90th percentile of the absolute pip size of losing positions for one signal and one symbol, and round-trip cost is the measured spread plus commission converted to pips as the median of commission divided by profit per pip. The four scripts and their printed output are kept with my working notes and are available on request.