Transparent rules. No black box.

Strategy settings.

Configure the deterministic ceiling, floor and hammer engine before testing.

Pattern & safety filters

Long wick / body minimum

Hammer or rejection wick must be at least this multiple of the real body.

Opposite wick / body maximum

Maximum short-side wick relative to the real body. Tiny-body doji are excluded.

Level tolerance (fraction)

0.0006 = 0.06% of price. Used for swing clustering and proximity.

Swing confirmation bars

Strictly lower highs or higher lows on both sides; right-side candles must have closed.

Strong breakout body / range

Strong bodies crossing a level return NO TRADE instead of fading the breakout.

Opposing-level clearance (ranges)

Minimum room to the opposing level in multiples of the signal candle range.

Maximum candle age (seconds)

Age measured from the most recent completed candle close. Hard cap: 90 seconds.

Entry window after close (seconds)

Directional signals only within this window, with the newest close present. Hard cap: 10s.

Require 5m and 15m trend agreement

Only fully completed, contiguous higher-timeframe groups are used. Disabling this filter does not bypass freshness or breakout protection.

Forex data provider

Server-side Forex adapter

Massive requires MASSIVE_API_KEY with current Currencies data entitlement in Cloud → Secrets. Current candle timestamps are checked; delayed or missing candles block live signals. Twelve Data requires TWELVE_DATA_API_KEY and is explicitly delayed. Changing provider does not move your journal or settings storage.

Performance modeling assumptions

Enter your demo environment's payout and estimated all-in transaction cost. Values are units per one unit staked, not an actual payment.

Net win payout / stake

0.80 models 0.80 units net gain per win. A loss is −1 unit.

Cost per resolved trade / stake

0.01 deducts 0.01 units on every resolved trade, including draws.

The app name includes AI, but this strategy is a deterministic, inspectable rule engine—not a trained predictive model. Confidence is an uncalibrated heuristic, not a measured win probability. Saving settings never changes existing recorded decisions.