Performance page and the walk-forward backtest
Running a backtest with the same engine as the live signals, reading the report, the honest walk-forward figures and the live signal tracker.
Why backtest first
The Performance page is the proof page. The backtester replays confirmed patterns from the stored history through the same signal fusion and trade-plan builder that produce live signals, with fees, slippage and leverage modelled. Nothing is simplified, so if a pair does not make money here it will not make money live. Check it before trusting a signal and again before switching an exchange account from Paper to Live.
Running a backtest
In the Backtest section:
- Pick Symbols from your watchlist (the first three are preselected) and Timeframes (default 1h and 4h).
- Set start and end dates (default: the last 180 days to now).
- Optionally type a threshold (
0.75or75; blank uses the global setting), leverage (default 5), risk % of equity per trade sized from the tight stop (default 2), and start equity (default 1000). - Click Run backtest. A progress bar shows
backtest #idand the percentage; the report appears when it finishes.
Previous runs are listed under Previous backtests with their parameters and status; click one to reopen its report. In the hosted service subscribers can view backtests; starting new ones is an admin action because the run uses engine resources shared by everyone.
Reading the report
The tiles give trades, win rate, return on starting equity, max DD (peak-to-trough drawdown), profit factor (gross profit over gross loss), expectancy (average P&L per trade), avg win / loss and equity start to end.
| Section | What it shows |
|---|---|
| Walk-forward (honest) | Trades, win rate, expectancy, return and drawdown when the confidence threshold is re-chosen every month from the previous three months only. No future information leaks into these numbers; they are the ones to believe. |
| Equity curve | Account equity after each closed trade (solid) with the walk-forward curve dashed in blue and the starting equity as a reference line. |
| Breakdown | Results grouped by pattern, tf, symbol or direction: count, win rate, average and total P&L, expectancy. |
| Threshold sweep | The same trades re-filtered at different confidence thresholds. The bar chart and table show count, win rate, average and total P&L per threshold; the suggested badge marks the threshold that maximised total P&L with enough samples. |
| Trades | Every simulated trade: entry and exit time, symbol, timeframe, pattern, direction, confidence, prices, unleveraged P&L %, P&L $, bars held and the exit reason. |
No trades. with a message usually means the date range holds no confirmed patterns on the chosen timeframes; widen the range or run deep analysis first.
In-sample versus walk-forward
The headline tiles use one threshold for the whole range, which is chosen with hindsight. The walk-forward block re-selects the threshold month by month using only the past, which is what you could actually have done. Expect it to be worse. If the walk-forward expectancy is negative while the in-sample one is positive, the pair's edge is not robust.
Exit policies and calibration
Two further sections sit under the backtest: Exit policies (which way of managing the trade paid best on this pair) and Calibration with the Auto-adjust threshold switch. They are described in Exit policy backtest and Confidence threshold and auto-calibration.
Live signal tracker
The bottom section summarises real signals, not backtests: those that resolved as target hit, stopped or expired. Tiles show resolved count, win rate, average and total unleveraged P&L, and the crash alert hit rate per level with the number of alerts and median lead time. Three tables break the signals down by pattern, timeframe and symbol.
Compare the live tracker with the backtest for the same pairs. A live win rate far below the backtest after 30 or more signals means the market regime changed or the engine needs a retrain; both call for caution, not a lower threshold.
Note: the backtester works on confirmed pattern signals. The consensus card has its own live track record on the Techniques & combinations tab and in the Calibration section; it is not part of the backtest report.