Analysis and Labs
Monte Carlo, walk-forward and nudge tests tell you whether a result is a plateau or a lucky peak.
Backtest on a fleet of workers, stress-test the results, build portfolios and hold every live account to its backtest. It all runs on your machine, and Claude, Codex or your own agent can drive every feature.
Windows desktop app for MetaTrader 5.
Backtests in one window. Results in a spreadsheet. Live accounts in another app. Your AI in none of them.
Every panel is live. Every panel is a tool your agent can call too.
Monte Carlo, walk-forward and nudge tests tell you whether a result is a plateau or a lucky peak.
A score per EA, tracked over time. Know when to pull one.
Check the fleet and your accounts from your phone or another computer.
Queue it once. Parallel MT5 workers do the rest.
Spot clones in the correlation matrix, then pick a mix off the Pareto frontier.
One config block gives your agent 200+ tools.
{
"mcpServers": {
"uas": {
"command": "uas-mcp"
}
}
}
Each step hands its output to the next, so a backtest you ran in March is still attached to the EA you deployed in June.
Broker symbols and tick data managed in one place, next to your own EAs, set files and custom symbols.
Queue backtests and optimisations from one terminal. A fleet of MT5 workers runs them in parallel while you do something else.
Detailed analytics with suggestions, plus robustness Labs: Monte Carlo, walk-forward, nudge tests, “plateau or lucky peak?” and “is this set overfit?”
Combine sets, read the correlation matrix, and choose allocations along a Pareto frontier of return against drawdown.
Journal every deployment. Each EA carries a live-vs-backtest score over time, so you see drift before it costs you.
And it bends to how you work: themes, layouts, saved views, custom Labs and skills.
UAS splits every job into passes and fans them out across a fleet of MT5 workers on your machines. You watch one queue instead of twelve terminals.
$ uas backtest submit \
--ea "Halcyon Gold" --sets v3..v8 \
--symbols XAUUSD,XAGUSD \
--from 2019-01-01 --to 2025-12-31
Run Halcyon Gold v3 to v8 on gold and silver since 2019.
uas_batch_submit 12 jobs, 2 symbols
Not a report file in a folder. Each backtest opens into metrics, an equity curve you can inspect trade by trade, monthly returns, the parameters that made it and the Labs already run on it. Pick another result from the library to compare.
v14, XAUUSD H1
The best pass in an optimisation is often a spike: one cell that did well while its neighbours didn't. UAS maps the whole surface, scores each pass by its neighbourhood, and turns the pass you choose into a new set version.
| Pass | Period | Stop | Net | PF | Neighbours |
|---|
Robustness tests come packaged as Labs, so you, the CLI or an agent run them the same way every time. Every verdict lists what was measured and, just as clearly, what wasn't.
Tick EAs in and out and the combined curve, metrics and correlation matrix update as you go. Then pick an allocation straight off the Pareto frontier of return against drawdown.
| In portfolio | EA | Weight | Net alone | DD alone |
|---|
Point UAS at an MT5 account and it pulls every deal, rebuilds the balance, splits the result by EA and holds each EA to the backtest you deployed it from.
9 closed, +$318, 6 opened
2,318 deals since 2025-01-06. Credit $0.00 kept apart.
| Expert advisor | Symbol | Trades | Win rate | Profit factor | Net | Live score |
|---|
| Closed | Expert advisor | Symbol | Side | Lots | Entry | Exit | Slippage | P&L |
|---|
One line per EA. The selected EA is drawn bright.
Go-lives, set changes, broker moves and pulls are logged as they happen, by you or your agent. When a score moves, the reason is already written down next to it.
UAS replays each live trade against its backtest and scores how closely they agree. When the score crosses your line, you hear about it with the evidence attached.
Balances are rebuilt from the deals and checked against the broker's number. Any difference is shown, not smoothed over.
Every go-live, set change and broker move is logged, so a score change always has a reason next to it.
Your accounts, ticks and results live on your PC. Remote access is opt-in and per device.
UAS maps tick and bar coverage for every broker symbol, month by month. A missing month shows up as a gap, so a quiet backtest never hides an empty tick file.
Gap, not zero. XAUUSD on Kestrel Prime is missing 6 trading days in July 2025. Backtests that cross it report the gap instead of treating those days as flat.
Every screen in UAS is backed by the CLI and an MCP server with more than 200 tools. Point Claude, Codex or your own agent at it and ask in plain language.
Is the new Meridian Breakout set overfit? Check it before Monday.
uas_set_show Meridian Breakout v14, XAUUSD H1
uas_lab_run_start lab: overfit-check, 400 nudges
uas_robustness_run walk-forward, 6 windows
It holds up. 83% of nudged sets stay profitable and walk-forward efficiency is 0.71. One weak spot: the Asian session filter. Removing it drops profit factor from 1.41 to 1.12, so that parameter is doing most of the work.
I've added a journal note and queued a 3-month forward test on worker 07.
uas_journal_add
These are the kinds of requests people hand to an agent connected to UAS. Every step is an MCP tool call you can read, rerun or do by hand.
Remote access is off until you turn it on, and it is opt-in per device. Link a phone or a laptop with a one-time code from the desktop app, then revoke it with one click. The data stays on your PC; the phone only views it.
Measured nothing is never reported as found nothing.
If a test couldn't run, or the data wasn't there, UAS says so. A blank is never dressed up as a clean result.
Every figure links back to the trades and deals it came from.
Nothing is screen-only. If you can click it, an agent can call it.
No cloud copy of your strategies. Nothing leaves unless you send it.
UAS is still being built with its community. Try it free and tell us what's missing.