Session is a session-channel trading system for MetaTrader 5.
One channel per session. One position set per channel. Nothing else, ever.
Beside the EA runs a second process. Every five minutes it reads the economic calendar, applies fixed rules for high- and medium-impact events, reads the headlines, and asks a local model one question: does this look worse than the rules say?
It can tighten. It cannot loosen. If it goes quiet, the EA trades on its own rules. Every stance it takes is written to a ledger you can open.
Two runs of the same EA on GBPUSD M15, September 2025 to August 2026. Read net and profit factor before anything else. This is a system that wins often and loses small; it is not yet a system that nets profit, and this page will say so until the sweep changes it.
| Run | Model | Trades | Win rate | Max drawdown | Net on $10,000 | Profit factor |
|---|---|---|---|---|---|---|
| v3 · defaults | Every tick | 514 | 73.74% | 4.68% | −$97.31 | 0.97 |
| v4 · gate off | 1-minute OHLC | 499 | — | 7.43% | −$521.20 | 0.86 |
| v4 · gate on | 1-minute OHLC | 493 | — | 7.27% | −$501.87 | 0.86 |
Average win $8.35 · average loss −$24.17 · zero rejected orders · zero errors in 4,242 lines of tester journal. The shape to fix is the payoff, not the entries.
Before the first backtest, the code went through a full review against the strategy as written. What it found, and what was fixed, is listed. Nothing on this page comes from a screenshot.
The EA runs inside your own MetaTrader 5 terminal, on your own broker account. There is no server of ours in the loop and no deposit to make with us.
The Specialist writes each stance to a ledger. The EA writes each trade to the terminal journal. You can read both. So can anyone you ask to check our work.
If the Specialist stops, the EA keeps trading its own rules. If the terminal restarts, the channel is rebuilt. If Friday comes, everything is flat.
Same EA, same year, gate on and gate off: profit factor 0.86 both ways. The gate fired 357 times and withheld six entries. Verdict written the same day: news timing is not where this strategy loses money. The losses come from the payoff shape, winners near $9 and losers near $25. The next lever is a TP1, TP2 and stop sweep, not more filtering.
A Python service next to MetaTrader, not inside it. Calendar, fixed rules, headlines, a local model that may only tighten, one line written to the terminal's common folder every five minutes. The EA reads it each minute and defaults to trading if the line is missing or stale. Forward test started on a demo account the same day.
514 trades, 73.74% won, 4.68% max drawdown, net −$97 on $10,000. The code held: weekend close every Friday at 17:45, re-entry gating as designed, no stuck positions. The one real finding was that a win pays about a third of what a loss costs, which is a strategy conversation, not a code fix.
Session is an expert advisor for MetaTrader 5. It marks a channel from the session's opening range on GBPUSD M15, trades the break of that channel with a stop and two targets, closes half at the first target, allows one position set per channel, and goes flat before every weekend. A second process, the Specialist, reads the news calendar and can stand the EA down around high-impact events.
Not yet, and the numbers section says so. In a year of backtests it wins about three trades in four with drawdown under five percent, and still nets slightly negative because losses cost about three times what wins pay. The next work is a sweep of the target and stop distances with the strategy's author. We will publish that run here whichever way it goes.
All of the usual ones. Past backtests are not future results; the tester's calendar grades more events as high-impact than the live feed does, so the live gate is looser than the tested one; a broker's fills and spreads differ from the tester's. Run it on a demo account first, for weeks, and then with the smallest lot your broker allows. Never with money you need.
A Windows machine with MetaTrader 5 and a broker account that offers GBPUSD. The Specialist additionally needs Python and a local model runner; it runs on a consumer GPU and needs no API key. If you skip the Specialist, the EA runs alone.
Write to us below. You get the compiled EA, the Specialist, the install notes, and the same backtest reports this page is built from. Pricing is settled in that first reply.
Everything on this page, in more detail: the strategy as written, the review findings, the full tester reports, and the Specialist's ledger format.
The EA, the Specialist, install notes and reports. Pricing in the first reply.
Every number on this page traces to a report file. Ask for the files.
Dated entries, including the ones where the idea didn't work.