Software Updates

The prop firm tab had been reachable only by adding a parameter to the address, which meant the flag had to be present on every page load: bookmarking the app, refreshing, or returning the next day removed the tab with nothing to explain why. It is now open to all accounts. The beta label stays, because the engine still has rough edges that are described below rather than hidden.

It answers two questions that need different simulations. Whether a strategy would have cleared a one-shot evaluation, with the profit target, daily loss limit, trailing drawdown and minimum trading days chained across every phase. And whether a funded account would have survived a year, where there is no target and no deadline and the only question is whether the drawdown rule is breached. Presets were included for FTMO, The5ers, FundedNext, FundingPips, E8 and Topstep, with custom rules for anything else.

On one run, 996 of 1,000 simulated paths passed every enabled rule. Reported as a pass rate that is 99.6 percent, and that is the figure most tools would print. Edge Matrix reported 69.8 percent.

The reason is that resampling a single backtest a thousand times measures how much the result depended on the order trades arrived in. It does not create a thousand independent observations. That record contained roughly seven independent challenge-length windows, so the reported figure is the 95 percent lower bound on seven observations rather than a count of a thousand replays of the same one. The result is separately discounted for edge decay before survival is simulated, using the record's own statistical significance and the proportion of calendar years it covered profitably.

The consequence is a lower number than other tools produce on the same file. That is the intended direction. Someone deciding whether to pay an evaluation fee is better served by a figure that understates than one that flatters.

Two things were making it slow, and neither was the simulation count the user chose. The persistence blend ran the full Monte Carlo sixteen times for a single result, so selecting a thousand paths quietly triggered seventeen thousand. Measured on a deliberately marginal strategy, where the blend genuinely moves the answer, sixteen draws against four changed the reported score by two tenths of a point on a hundred point scale while costing two and a half seconds. That is smaller than the run-to-run variance of the simulation itself, so the draw count was reduced to five.

The block sampler was also walking each simulated path one step at a time in Python, and it is called once per path, which came to several million interpreted steps per request and about two thirds of the total time. It was rewritten to compute the whole walk in one array operation. The two versions were checked against each other across three thousand random pool sizes and sample lengths, and against the same random seed at five pool sizes, and produced identical output every time. Challenge mode now completes in around a second where it previously took five.

The tab previously opened on a single line of grey text asking for a backtest. Someone arriving on it was told nothing about what it does, that there are two different questions it can answer, or why its figures sit below what other tools report. It now opens on the two modes side by side, phrased as the questions they answer, with a short explanation of the evidence bound and the decay discount underneath. On shorter screens the explanation is dropped rather than allowing the panel to scroll, since the mode choice is the part that matters.

The projection tab gained two controls. Simulation count could be set between one thousand and ten thousand paths, and the resampling method could be chosen rather than fixed. A line under the toolbar reported which method actually ran and, for block bootstrap, the mean run length that was read from the trade record. The exported report carried the same line, so a document opened months later named its own method instead of leaving the reader to assume defaults.

Three methods were offered. Block bootstrap lifts runs of consecutive trades whole, so losses that bunched together in the record can still bunch together ahead. Bootstrap draws single trades independently. Permutation reorders the exact set, so the final balance is identical every run and only the path changes. A diagram of each was added behind the help icon, showing the same seven trades mapping into one simulated run.

None of them is presented as the safe choice, because on measured data that claim does not hold. On a record with a 74 percent win rate, block bootstrap reported an 18.1 percent drawdown at the 95th percentile against plain bootstrap's 26.9 percent: drawing trades independently strung together losing sequences the strategy had never actually produced. On a strategy whose losses genuinely cluster, the comparison runs the other way. The difference between two methods on the same record is itself a reading worth taking.

Raising the ceiling to ten thousand paths exposed two problems that had been present at the old fixed count and simply cost less there. Drawdown recovery lengths were measured by walking each simulated path one trade at a time in Python, which at the largest settings meant ten million interpreted steps per request. Every simulated curve was also being retained at full resolution although every consumer of them reduced each to two hundred points first.

Recovery detection was rewritten to operate on arrays, and curves are now thinned as they are produced rather than afterwards. A ten thousand path projection over one thousand trades fell from a hundred and nine seconds to fourteen, and peak memory from a hundred and sixty megabytes to a hundred. The definition of a recovery did not change, which was checked by running the old and new implementations against four thousand random equity paths and confirming they agreed exactly.

The app could be switched between the dark interface and a light one from the tab bar, and the choice was remembered between visits. Every tab was covered rather than a subset: charts, correlation grids, test panels and the modals over them all changed together, because the switch was applied at the root rather than panel by panel. Screenshots and logos kept their own colours, since a photograph has no light-mode equivalent and inverting one only makes it wrong.

Nothing about the dark interface changed. The light rules apply only when the theme is set, so turning it off returns the app to exactly what it was before. Colours in light mode were derived from the dark palette rather than picked individually, which covered the whole application at once and is honest about being approximate in places rather than hand-tuned everywhere. Exported reports were not affected by the switch and kept their own styling.

Any report could be published to a link that anyone could open, with no account and no sign-in. The published copy was frozen at the moment of sharing, so the figures a link showed on the day it was posted were the figures it showed a year later, and it kept working after the session that produced it had ended. Each published page carried a line stating that the report was self-published, because a report chosen by the person selling the strategy is selected evidence and a reader who has to work that out for themselves discounts everything else as well. Links could be withdrawn at any time from a list in the app, which also showed how many times each had been opened.

Every report carries a reference, and that reference could be looked up on the site to see the strategy, date and headline figures recorded when the report was published. A figure in a circulating copy that did not match had been changed after the fact, and an unknown reference had never been published. Three things kept the lookup from becoming a way to read other people's results. It answered only for reports whose publisher chose to make them checkable. Unknown, malformed and unlisted references all returned the same answer, so nothing about the response revealed which references existed. And the reference itself gained a random suffix, since the previous format was a plain timestamp with 86,400 values a day and could have been swept.

Withdrawing a link stopped the page loading but left the reference resolvable, still reporting what had been published. Content stayed under the publisher's control while the record of it did not, so a report shown to someone and later withdrawn could not afterwards be denied.

Two phrases described a backtest rather than reported it. A Calmar ratio above the reference range read as significantly exceeding the professional benchmark, and a favourable penance ratio read as substantially outperforming the theoretical expectation, with the addition that this was a strong signal of genuine edge. Both now state where the figure sits and stop there. An unusually high ratio is the kind of number that warrants an explanation rather than congratulation, and a report that argues for the strategy it is measuring is worth less to the person reading it.

Edge Matrix read the spreadsheet that cTrader produces from the History tab, alongside the HTML strategy tester report it already accepted. Each position's net result was taken directly, so commission and swap were counted once rather than added twice, and dates were read day first to match cTrader's own format instead of being guessed at. The parser was checked against the HTML report of the same account and returned the same 207 trades in the same order for the same total. One difference mattered: unlike the HTML report, the spreadsheet carries no running balance, so the account's starting balance was asked for at import. Percentages are measured against it, and on one test file assuming the wrong figure turned a 5.95 percent drawdown into 0.77 percent. Importing without supplying it still worked, and the dashboard then stated on the result that the percentages rested on an assumed balance.

The separate CSV and MT4/MT5/cTrader buttons became a single area that took any supported file and worked out what it was from the file itself. Files could be dragged onto it as well as picked through a dialog, and a file chosen through what used to be the wrong button reached the right parser regardless. The accepted formats were listed on the control rather than left to be discovered: MT4, MT5, cTrader, CSV and XLSX.

Every new account received one full validation at no cost, with no payment method asked for at any point. It was not a demo or a reduced version: it ran all 20 statistical tests, the Edge Matrix Score, Monte Carlo at 50,000 iterations, Insights, Compare, Portfolio Mode and all 8 exportable reports on a backtest of the account holder's choosing, and produced exactly what a paid validation produced.

Validations became purchasable in one-time packs rather than only through a subscription. A pack of 5 cost $9.99, 25 cost $29 and 75 cost $59, and one credit covered one new backtest. Packs were one-time purchases, so nothing renewed and nothing needed cancelling. Subscriptions continued unchanged for anyone who preferred unlimited validations, and a subscriber's credit balance sat untouched while the subscription was active.

A credit was charged the first time a particular backtest was validated, and never again for anything done with it afterwards. Every test, every Monte Carlo run, every Insight, every Compare, every Portfolio and all 8 report exports on that backtest were unlimited and free from that point on, including loading the same file again. A backtest was identified by its trade data, so a re-export of the same result was recognised as the same strategy, while re-running the same expert advisor over a different period produced different trades and counted as a new backtest.

The tab bar showed the remaining balance for accounts using credits, and hid it entirely for subscribers and Lifetime holders, who were never metered. Clicking the balance opened a panel listing every backtest the account had already paid to validate, with the instrument, timeframe and the time it was validated, each marked as free to load again. Importing a backtest showed whether a credit had been spent or whether the file was already owned, rather than leaving a balance change unexplained.

Portfolio Mode needs at least two validated strategies to produce anything. Selecting more files than the remaining balance covered previously spent a credit and then left a portfolio that could not run. The app now checks before uploading anything, explains what the balance covers, and spends nothing when the result would be unusable. Comparing strategies already validated cost nothing, and a comparison that could not be afforded was refused before any credit moved rather than partway through.

The pricing page was rewritten around the choice between paying per strategy and paying for unlimited use, with the free validation shown first. The Solo Trader plan was withdrawn from sale; existing Solo subscribers were unaffected and kept the plan and the access they had. Refund terms were updated to cover credit packs, with unused credits refundable within 14 days of purchase at the pack's per-credit price and credits already spent treated as delivered.

Edge Matrix read position exports downloaded from MQL5 signal pages, alongside the MT4 and MT5 account statements it already accepted. These exports record deposits and withdrawals in the same list as the trades, so those entries were separated out rather than counted as trades, which would otherwise have added a fake winner the size of each deposit. Each trade result combined profit, commission and swap so the figures matched the account, and performance was measured against the capital that was actually at risk at the time of each trade rather than against the full deposited total. Signal accounts that traded more than one instrument were recognised as such instead of being reported as a single pair. The CSV import option was enabled on the import screen for these files.

When a simulation was run with block bootstrap, the exported report described it as a plain bootstrap. The result itself was always computed with the method you selected, but the report labelled it wrongly. The report now names the method that actually ran, and older saved results that carry no method setting still read correctly.

The Monte Carlo section of the exported report now explains the resampling method that was used and why it was appropriate. For block bootstrap it sets out how runs of consecutive trades were resampled so that losing streaks and volatility clustering survived into the simulated paths, and it shows the figures behind the choice: the automatically selected mean block length, the lag-one autocorrelation of trade results, the same measure on squared results, and the significance threshold each was judged against. Block length was selected from the data rather than set by hand, and the selection ran on squared results because drawdown is driven by clustering in the size of moves rather than by dependence in the average trade. Bootstrap and permutation runs carry their own shorter explanations.

The full report, the one that covers analysis, Edge Matrix, Monte Carlo and the closing verdict together, now carries the Statistical Plausibility result throughout rather than only in the test list. It appears as its own test, the executive summary and the closing recommendation adapt when a strategy is flagged, the conclusion tally reports the gate separately from the averaged tests, and a technical section sets out the mathematics with the strategy's own figures. A flagged backtest no longer reads as an endorsement anywhere in the document.

Edge Matrix added a validation test that asked a question the others could not: was the combination of win rate, reward-to-risk, and trade frequency even consistent with a genuine, durable edge? Using the relationship between those quantities and the Sharpe ratio (Lopez de Prado, Advances in Financial Machine Learning), the test calculated the annualised Sharpe a strategy's results implied, then compared it against the highest verified track record on record. Results beyond that bound were flagged as more consistent with overfitting or performance that may not persist out of sample than with a durable edge. The full calculation was shown in the app and the exported report, every input, the formula, and the sources, so it could be audited directly. It was applied as a score gate rather than averaged, so a smooth-looking overfit curve could not outvote its own detection. Genuine strategies passed untouched. The test measured statistical plausibility; it was not a determination of intent.

When the plausibility gate engaged, strategies were no longer collapsed toward a single midline score. Each was mapped into a capped band whose position preserved its ranking against other gated strategies, so a strategy that was only implausible sat above one that was implausible and weak across the other tests. Gated strategies remained clearly below a genuine pass, but they spread by real quality instead of flattening to one number.

The exported Edge Matrix report carried the new test as its own card, led the analytical findings with it when a strategy was flagged, and included a full technical section, the formula, the strategy's own figures, and the research basis. The report abstract adapted to the result: an implausible strategy no longer read as an unqualified endorsement, while a genuine one stayed clean.

Statistical Plausibility led the validation test list. Each test's detail panel opened its own explanation reliably, and an information button revealed the underlying mathematics and research for the plausibility test. Redundant risk-findings text that duplicated the new test was retired, so a single consistent explanation remained on screen.

Portfolio analysis now required at least two strategies before it ran, since a single strategy is not a portfolio. Loading one on its own returned a clear prompt rather than an empty portfolio result.

The free Monte Carlo analyser accepted trading-journal exports in addition to MT4, MT5, and cTrader reports. It located the profit column by name, read the per-trade results, and ran the same browser-based simulation. Everything stayed local to the browser, with no upload.

Edge Matrix now read live and demo account statements, not only strategy tester backtests. MT4 account statements and MT5 trade history were parsed directly, with deposits, withdrawals, and transfers separated from the trading. For an account funded over time, performance was reconstructed against the balance that was actually at risk on each trade, so drawdown and returns matched how a broker or a track-record service reported the same account rather than assuming all the capital was present from day one. The dashboard showed absolute gain, the true maximum drawdown, and the time-weighted return. The full statistical suite ran unchanged, and the integrity checks stayed active, since a statement could be incomplete or unrepresentative just as a backtest could. Report wording adapted to the loaded account history instead of referring to a backtest.

The Monte Carlo tab now lets you choose your resampling method, Bootstrap (resample with replacement), Permutation (reorder, no replacement), or both side by side. "Both Methods" ran the same backtest under identical settings and compared the results, which isolated how much the robustness score depended on the method rather than the strategy. The exported report included the full comparison, with a separate robustness grade for each method.

A new eye toggle in the tab bar replaced your strategy's real name with an anonymous code everywhere, across all tabs, reports, and download filenames. Set it before loading and the name never appeared on screen, which suited sharing screenshots or recording videos without revealing the strategy.

Projections moved to block-bootstrap resampling, which preserved winning and losing streaks instead of treating each trade as independent. Confidence intervals were added on key metrics, so the numbers reflected how returns actually clustered rather than an idealized random walk.

Corrected how trading days were pooled, which had compressed the timeline for strategies that trade less often. Projections then spanned the full backtest date range honestly instead of overstating activity.

Cleared up faint or hard-to-read labels in several panels, aligned scrollbars to the app's theme, and tidied a range of small interface details across the tabs.