Questions & troubleshooting

A Windows PC, a MetaTrader installation, and the KB Algo Lab desktop app from the download page. That's it — the app drives your own copy of MetaTrader, so there's nothing to configure on the broker side.

No. Backtesting runs entirely against historical data on your own machine. A demo account is enough to have MetaTrader installed and working.

MT5 is fully supported and is what we'd recommend. MT4 works with a portable-style install (where the MQL4 folder sits beside terminal.exe). If you have a choice, use MT5 — its strategy tester is the one built for real tick modelling.

On your account page once you've signed in. Paste it into the app's Share tab. The key is tied to your account — keep it to yourself, as a key found in use by several people can be reset.

No. Pro and Elite stream it directly from our server inside the app — 18 symbols, roughly 27 GB in total, already built and ready. If your connection drops mid-download it resumes where it left off rather than starting again.

Eight major FX pairs, nine crosses, plus gold and silver — 18 in total. Every one runs from 15 July 2024 to yesterday, a rolling two years.

The server tops the data up automatically every hour, so it stays current without you doing anything.

Genuine bid/ask tick history with the real variable spread as it moved — not M1 bars resampled to look like ticks. That distinction is the whole point: a fixed-spread approximation flatters scalping strategies enormously.

Your chosen window falls outside the data you actually have. It's the single most common cause of a run that won't start, and the app now tells you the exact coverage it holds instead of failing vaguely.

This is the one we've seen most. Nearly always the MQL5\Experts folder isn't writable — which is what happens when MetaTrader is installed under C:\Program Files, where Windows blocks writes from ordinary programs. The .set file lands fine because it goes somewhere else, which makes it look like an EA-specific problem.

Same root cause as above — the optimiser couldn't write its job file into the MetaTrader folder. Updating to the current version resolves it, as the app now resolves the right data folder for both standard and portable installs.

MetaTrader allows only one terminal per data folder. If a terminal is already open on that folder, a second launch quietly hands over to the first and exits without running anything — so the app waits for a result that never comes.

The results panel now lists the specific reason for each one rather than just reporting a count. Read those first — the usual causes are a date window outside your data, a symbol you haven't built data for, or MetaTrader being open.

Yes. The app launches its own MetaTrader instance to run the test. If you already have one open on the same data folder, the run can't start. Close it and let the app do the launching.

Email hello@kbalgolab.co.uk with the exact error text and what you were doing. The precise wording usually identifies the cause immediately.

Because we charge the costs most vendor backtests quietly omit: the real variable spread from bid/ask ticks, broker commission, and measured overnight swap. In our own testing, adding commission and swap alone turned roughly 30% of previously profitable results into losers.

That gap isn't a fault in the test — it's the part that decides whether a strategy survives contact with a real account.

It reshuffles your trade sequence thousands of times and shows the resulting spread of outcomes. A single backtest is one ordering of your trades and can flatter badly. If the strategy only looks good in the order it happened to occur, this is where that shows.

The window is split: the strategy is judged on the first part, then re-checked on a later part it was never tuned against. Strong in-sample numbers with weak out-of-sample ones are the classic signature of curve-fitting — settings tuned to noise that won't repeat.

Spread. On very low timeframes, trade counts run into the tens of thousands, and the spread paid on every one of them overwhelms the edge. In our own sweep of rule-based strategies, M1 was profitable in 0 of 180 combinations tested, against roughly half on H4 and D1.

It isn't a bug, and it's worth knowing before spending an evening optimising there.

The account was mathematically wiped out during the run. Treat those as failures regardless of how the final profit figure looks — the strategy would have been closed out long before reaching it.

It analyses strategies and results, and can run real backtests on your own tick data to answer a question rather than guessing. It works from measured findings and will say when it doesn't have the data instead of inventing a figure.

It is not a financial adviser: it won't tell you what to buy or predict prices.

Free — one pair, three timeframes, modelled data. Enough to see how the app works end to end.

Pro — £19/month or £209/year (a month free). Unlimited pairs and timeframes, real bid/ask tick data, and leaderboard publishing.

Elite — one-off £1,500. The same features permanently, plus priority in the cloud queue.

Yes. Cancel from your account page at any time through the Stripe billing portal. You keep Pro until the period you've already paid for ends, then drop back to Free — nothing is deleted.

You can request a full refund within 14 days of purchase provided the Elite features haven't been used. After that it's non-refundable. Email us inside the window and we'll sort it.

No. Free is genuinely free and doesn't expire. It's limited to one pair and three timeframes so you can judge the app before paying for anything.

Not for local backtests. Your EA files, .set contents and backtest history stay on your own machine. They only leave your computer if you deliberately use the Cloud Runner, ask the Assistant about a specific file, or publish a result to the leaderboard.

Your account email, licence key and plan, plus anything you've explicitly chosen to publish. The privacy policy sets out the detail.

Only runs you publish yourself. Published results carry an honesty tag showing whether they came from real ticks or a faster modelled run, so a flattering fast run can't be passed off as a real-tick result.