Muninn

A new robot for MetaTrader 5 driven by a neural network — an LSTM, a model with memory — that decides both the entry and the exit on gold, on 15-minute candles. It holds one position at a time, with a real stop loss in the order from the first second. No grid, no averaging down, no martingale, ever. What it has today is a promising test and no live record.

In measurement Zero live operations High-risk instrument MetaTrader 5

Platform: MetaTrader 5MQL5, Windows terminal
Market: GoldXAUUSD, 15-minute candles
Model: Neural networkAn LSTM decides entry and exit
Live operations: 0Running since 20 September

Live record

Zero operations. None.

Muninn has been running since 20 September on one of our own demo accounts and has not opened a single position. Its signals have not reached the confidence it requires to enter.

This is not a footnote: it is the state of the product. There is no live curve to show, no net, no drawdown measured in real money. A robot that has not traded has no track record, and lending it one from the tester would be exactly the trick these pages exist to avoid.

What it is

A new robot, not a version of Odin.

Muninn is Odin's raven, the one called "memory", and the name is the only thing it really inherits. From the robot Odin it borrows the shell — the licence, the logs, the daily brake — and none of its strategy: the whole decision is the model's.

01 — THE MODEL

A network with memory decides

An LSTM is a neural network that carries what it saw before: it does not look at the current candle alone, it looks at the sequence. On 15-minute candles it emits a probability that the price will be higher a few candles ahead.

Above a threshold it buys; below the mirror threshold, it sells. Between the two it does nothing, and that is most of the time.
The exit is the model changing its mind. It closes when that probability crosses back, not on reaching a fixed target in points.
The model decides both ends. Entry and exit come out of the same belief; there is no human rule on top correcting its exit.

02 — THE SHAPE OF THE RISK

One position and a real stop

The geometry is the opposite of Goldy's, and that is deliberate: nothing here averages. One position, a stop placed in the order, and until it closes no other is opened.

One position at a time. It stacks no legs and holds no two sides open.
A real stop loss in the order from the first second. It travels on the broker's server: it is the catastrophe stop for when the model is wrong and the market gaps.
No grid, no averaging down, no martingale. Ever. It is not a setting you could change: that code does not exist.

What we measured

Strategy tester, flat lots, two windows.

MetaTrader 5 strategy tester runs at flat 0.1 lots on a 50,000 USD account. The second row is 2026 alone, on 100% real tick data, and it is the one carrying eleven operations.

Window	Profit factor	Maximum drawdown	Operations
January 2024 → 18 September 2026	1.79	1.99%	216
2026 alone 100% real tick data	1.24	1.43%	11

What these two rows do not prove

A strategy tester is not the market. It is a reconstruction of the price; slippage, spread widening and a rejected order do not hurt it the same way.

Eleven operations in 2026 is not a sample you can lean on. At that count, a single different trade moves the profit factor from one side of 1 to the other.

And a model measured on the same history it was chosen from will always look better than it is. That is the structural trap of any trained system, and it is not fixed by looking at the same table again.

What Muninn has today is a promising test and no live record. We would rather say that than dress it up.

Risk disclosure

Muninn is in measurement. Nothing on this page is an invitation to trade, and it is not investment advice.

It has no live record. Zero operations since 20 September on one of our own demo accounts. There is no real-money result to show.
Every number on this page comes from the strategy tester, on historical data, at flat 0.1 lots on a 50,000 USD account. None of it comes from an account holding customer funds.
The eleven operations of 2026 are not a sufficient sample to support any conclusion about future profitability.
A model trained and chosen on the same history measures in its own favour. That apparent edge is not a market edge.
It trades a high-risk instrument. Leveraged gold via CFDs. It can cost the entire capital in an account.
Nothing published here is investment advice or a promise of returns. No past result, measured or projected, guarantees a future one.
Trading leveraged CFDs carries a high risk of losing money rapidly, including the entire deposit.

Want to know when it has a record?

Write to us and we'll tell you when Muninn has traded enough for there to be something to read. Today there isn't.

trading@tuurt.com