Build a rule. Test the idea.
Combine pip distances, a signal, entry hours and exit rules. Compile the recipe, review what it means, then run a batch of tests.
Choose preset snippets
Each preset replaces its own rules and keeps the others.Original idea
Review the transferred fields and defaults below. Nothing has been compiled or backtested automatically.
Lookback and confirmation use 5-minute bars. A 12-bar lookback is 60 observed minutes. Tests counts scenario runs or chronological windows, not trades or optimisation attempts. Synthetic length supports 200–10,000 bars per test; historical tests use all imported rows. More synthetic rows do not establish real-market fidelity or provide independent historical evidence.
Scenarios or your historical bars
Real history is available through the local HistData downloader. Imported files appear in the dataset menu for their pair. Public tests use synthetic scenarios; provider files stay on your computer.
Download market history or connect an API
HistData offers free one-minute bid CSV archives for all four pairs. In the local project, run python3 local/fx_sources.py --pair EUR/USD --start 2023-10-28 --end 2026-10-28. Dates are UTC and end is exclusive. The downloader records source hashes, converts fixed EST to UTC and keeps only complete five-minute bars. Restart is not required; reload this page.
Frankfurter API supplies daily reference rates without a key. These suit long-term context, but have no intraday OHLC or executable spreads. Dukascopy offers historical exports subject to its use terms. The connected Massive plan rejected the requested 2023 intraday history. Free downloads do not establish public redistribution or model-training rights.
How do we recognise a trend or reversal?
Momentum compares the latest closed price with the close one lookback earlier, requiring the move_pips threshold. Reversal first requires that earlier move, then consecutive closing prices in the opposite direction. It enters at the following open. These are testable definitions of a candidate move; neither establishes that a new trend will continue. A trained trend-start predictor has not been deployed.
A four-hour demonstration is useful for mechanics checks. Market training needs longer source history, chronological training/validation/test splits, varied market conditions and evaluation after costs. Repeating one window or changing synthetic seeds cannot substitute for unseen real history.
Import a private five-minute CSV
Keep one pair per file, with header t,o,h,l,c. Use UTC millisecond start times and closed five-minute OHLC bars, oldest first. Gaps remain gaps. Maximum 32 MB or 400,000 bars per file.
Files stay in this local service. Importing does not grant training or redistribution rights.
Compiled strategy
No recipe compiled yet. The compiler checks every rule and shows any defaults.
Compile to see the validated settings.
Results after assumed costs
Run a compiled recipe to see results.Execution assumptions and interpretation
- Signals use closed bars. Entry occurs at the next bar’s open.
- These presets are hypotheses. They have not demonstrated a profitable edge.
Pip targets describe gross price movement. Net results subtract the full spread and slippage on both sides. One position at a time. No cash or margin limits, financing or broker execution guarantees are modelled. More tests do not automatically improve a strategy; keep a separate unseen period before choosing settings.