Our audits folder holds 38 files and five of them contain no run of their own. They hold the difference between runs. A comparison file is what sits behind the side by side matrix on Quantum Titan, Scalping Robot Pro, Smart Gold Hunter, The Gold Reaper and Quantum Queen X, and the interesting part of it is not the numbers it shows. It is the set of conditions under which it declines to show any.
Each file names its own axis. Two of them compare versions of one product, Quantum Titan across 2.1, 2.2 and 3.0, and Scalping Robot Pro across 3.0 and 4.0. Two compare presets of one build, the four Smart Gold Hunter settings and the two Gold Reaper settings. The fifth carries the axis cross product and compares two products whose runs turned out to be deal identical, which is a case of its own further down.
Between them the five files hold 13 audited variants, three plus two plus four plus two plus two. Every one of the 13 carries the status ok. None is waiting for a run.
The deltas in each file run against one named reference row rather than against an average of the others. For the version files that reference is the newest build, 3.0 for Titan and 4.0 for Scalping Robot Pro. Averaging two settings of one robot would produce a number that describes neither of them, so the matrix never does it.
Five fields have to agree across every variant in a file. Symbol, window start, run capital, leverage and the commission retrofit rate. If any of them differs the comparison is not built at all, it raises an error and names the two variants and the fields that disagree.
The reason is stated in the module itself. A comparison across two test setups measures the data, not the update. Put a run on a vendor supplied broker history next to a run on our own gold series and the difference you read is the history, whatever the version numbers say. That is not a hypothetical for this catalog, because four of our audits still sit on a vendor world with a different symbol suffix, and a version comparison against one of them would look perfectly reasonable on the page.
In all five published files the gate passed on identical values. Symbol XAUUSD, window start 2003.05.05, run capital 100,000 USD, leverage 1:5000, commission retrofit 3.50 USD per lot and side. The line above every matrix on the site states the same five values with the symbol written without its data suffix, which is the same claim in a form a reader can check.
The end of the test window is not a gate field. Versions get measured on the days they arrive, and demanding an identical end date would mean either rerunning every older build or refusing to compare at all. Instead the engine writes a note into the file, and the note is not optional.
Two of the five carry it. The Titan file records ends of 2026.08.17, 2026.08.19 and 2026.08.23 and states that absolute totals are not directly comparable. The Scalping Robot Pro file records 2026.08.17 and 2026.08.19 and says the same. The four Smart Gold Hunter presets and the two Gold Reaper presets all end on 2026.08.17, so those files carry no such note.
This matters for how the Titan matrix should be read. Version 2.1 audits to a net of 669.83 USD over 1,485 trades, version 2.2 to 488.07 USD over 1,503 trades and version 3.0 to 1,089.53 USD over 1,947 trades. Those are the audited net figures after commission and swap rather than the tester head. The file stores the gap between 2.2 and 3.0 as a delta of 601.46 USD. Six days separate the first window end from the last, and the note exists so that nobody reads that delta as the value of a version bump.
Three rows of the matrix depend on the deposit a vendor recommends rather than on our own run capital. Those rows appear only when every variant in the file carries the same vendor evaluation deposit. Where the deposits differ or one is missing, the rows are dropped and the file stores the reason together with the deposit of each variant, so the gap on the page has a stated cause instead of an empty column.
In all five published files that reason field is empty, which means all five show the capital rows.
The note list carries more than window mismatches. The Gold Reaper file records for both of its presets that the survival estimate is gated, because the prop firm map requires fixed lots while the measured lot policy came back as mixed or unknown. The corresponding cell in both columns is empty rather than filled with a number the gate does not support, and the reason for the emptiness is printed under the matrix.
A variant can be declared in the catalog before it has been measured. When that happens the row is built anyway with the status pending and a reason, and the page renders it greyed out without a link. The alternative would be a matrix with holes in it, which reads as an absence of findings rather than an absence of data.
No file carries a pending row today. The mechanism is worth knowing anyway, because the next version of any of these products will create one the moment it is declared and before it has run.
Four of the five comparison files end with an identity flag set to false, which is the ordinary case. Two runs of anything differ somewhere.
The Quantum Queen X file ends with that flag set to true. Every delta cell in it reads zero, across net profit, trade count, the end of day Sharpe, drawdown, the head profit factor, the days to 80 percent of the profit, break even shock, commission, swap, average lot and the three capital rows. The file also carries a role of republish forensics and a census note stating that our deal level census found 6,714 identical deals across both tester runs, with a net of 4,865.60 USD before the commission retrofit. Quantum Queen was delisted after Quantum Queen X launched, and the two tester runs are deal identical.
The flag is not a human judgement written into the file. It starts true and is switched off by the first non zero delta or the first differing verdict lamp. It also requires the capital rows to be present, so a file that lost those rows cannot claim identity even if everything visible matches. The comparison code carries 12 self tests for these paths, one of which rebuilds the Queen pair out of the repository and asserts the zero result.
It will not tell you that one version is better than another. It puts the verdict lamps of each variant next to each other without merging them, and it prints the window note when the runs are not strictly comparable. The lamps are visible to everyone, the number matrix sits behind the purchase, and the note travels with both.
It also will not compare across the thing that matters most and is hardest to see, which is the data underneath. That gate is the only part of the file that can stop the whole comparison from existing, and it is the part with no numbers in it at all.