Signal Gates

🦉 AthenaAI is an AI-powered ensemble machine-learning trading bot for PocketOption, built on BinaryOptionsToolsV2 → Get it on GitLab
💬 Need help? Join the Chipa Discord · ⚡ Go no-code with ChipaEditor

AthenaAI is deliberately reluctant to trade. Every iteration of _trade_loop in src/bot.py (running every poll_interval seconds, default 2.0) pushes the current market state through a fixed pipeline of gates, in code order. A single failed gate means continue — no trade this cycle. Only when every gate passes does the bot place an order (log line ▶ TRADE ...).

Blocked cycles are quiet by default: a diagnostic reason is logged at most once every 30 seconds (the ⏸ ... lines), so a silent bot is usually just a gated bot — see Troubleshooting.

The pipeline, in code order

Warmup — fewer than max(warmup_candles, 60) candles collected (default 60). The feature engine and SMA50-based features need history before predictions mean anything.
Max concurrent trades — len(pending_trades) >= max_concurrent_trades (default 1). One open position at a time by default.
Minimum wait between trades — less than min_wait_between_trades seconds (default 60) since the last order.
Daily loss limit — MoneyManager.can_trade() returns false once the day’s losses reach max_daily_loss (default $300). Trading stops for the day.
Cooldown — a previously triggered consecutive-loss cooldown (_cooldown_until, see gate 15) has not yet expired.
Trading hours — the current UTC hour is not in trading_hours (default (0,1,2,3,4,16,17,19,20,21,22,23); the v6 comment notes 08–15 UTC lost money consistently over a 546-trade analysis). An empty tuple disables this gate.
Adaptive cooldown — the adaptive strategy has imposed an extra cooldown (120 s or 300 s) after a cold streak, and not enough time has passed since the last trade.
Feature availability — FeatureEngine.compute() returned None (fewer than max(feature_window, 26) candles). The surviving vector is sanitized with np.nan_to_num.
Volatile-regime skip — regime is VOLATILE and skip_volatile_regime is true. Off by default (False in src/config.py — “let ML decide”).

The ensemble now predicts a direction and confidence; the remaining gates judge that signal.
Minimum confidence — confidence < min_confidence (default 0.63, PO_MIN_CONF). Also clears the signal-confirmation history, so a weak candle restarts any confirmation streak.
Maximum confidence — confidence > max_confidence (default 0.85, PO_MAX_CONF). Overconfident signals are treated as overfit noise. Also clears the signal history. (The ensemble already clamps confidence to 0.78, so this gate only fires if you lower the cap.)
Low-range candle — the latest candle’s high − low < 0.00005 (about half a pip on EURUSD). Doji/indecision candles produce near-random outcomes. Hard-coded threshold.

Indicator alignment — with require_indicator_alignment true (default), _check_indicator_alignment demands that at least 2 of 3 classic indicators agree with the ML direction:

Indicator (feature index)	CALL vote if	PUT vote if
RSI zone (14)	`rsi < 70` (not overbought)	`rsi > 30` (not oversold)
MACD histogram (12)	`macd_hist > 0`	`macd_hist < 0`
SMA cross, SMA5−SMA20 (10)	`sma_cross > 0`	`sma_cross < 0`

Note the RSI check reads feature index 14, which is the −1/0/+1 rsi_zone value rather than the raw RSI (index 13) — in practice the zone value always satisfies < 70 and > 30, so this vote effectively always passes and alignment hinges on MACD and SMA agreeing.

Signal confirmation — _check_signal_ready requires the last signal_confirmations consecutive candles (default 1 = instant) to agree on the same direction. One signal is recorded per unique candle timestamp; a repeat check on the same candle returns false (wait for the next candle). With N > 1 you’ll see 📡 Signal building: k/N candles agree on call and finally ✅ Signal CONFIRMED: call xN candles avg_conf=…%. Mixed directions reset the wait. History is cleared after a confirmed signal and after any confidence-gate failure.
Consecutive-loss cooldown trigger — if perf.consec_losses >= max_consec_losses (default 3), set _cooldown_until = now + cooldown_seconds (default 300 s), reset the counter, and skip. Future cycles then fail gate 5 until the cooldown expires. Log: Hit N consecutive losses → cooldown 300s.
5-minute multi-timeframe trend — if at least 100 candles exist, the loop samples every 5th candle (candle_list[::5]) to synthesize a 5-minute view, and when at least 20 such samples exist compares SMA5 vs SMA20 of the sampled closes. A CALL is rejected if the 5-minute trend is down (sma5_5m <= sma20_5m); a PUT is rejected if it is up. No config knob; skipped entirely with fewer than 100 candles.
Adaptive should_trade — the adaptive strategy’s final veto: blocked regime, blocked hour, non-preferred direction without +0.08 extra confidence, or confidence below min_confidence + confidence_adj. Log: 🧠 Adaptive skip: <reason>.

After gate 17 the bot sizes the stake (Kelly-based, see Risk management), asks the expiry selector for a duration, and places the order.

Summary table

#	Gate	Condition to pass	Config knob (default)	Diagnostic log when blocked
1	Warmup	`len(candles) >= max(warmup_candles, 60)`	`warmup_candles` (60)	`⏸ Warming up (n/60 candles)`
2	Max concurrent	open trades < limit	`max_concurrent_trades` (1)	`⏸ Max trades open (1/1)`
3	Min wait	≥ N s since last trade	`min_wait_between_trades` (60 s)	`⏸ Wait between trades (Ns left)`
4	Daily loss	daily loss < limit	`max_daily_loss` ($300, `PO_MAX_DAILY_LOSS`)	`⏸ Daily loss limit reached`
5	Cooldown	past `_cooldown_until`	`cooldown_seconds` (300 s)	`⏸ Cooldown (Ns left)`
6	Trading hours	UTC hour in whitelist	`trading_hours` (12 allowed hours)	`⏸ Outside trading hours (HH UTC)`
7	Adaptive cooldown	≥ extra s since last trade	learned: 120/300 s	`⏸ Adaptive cooldown (Ns)`
8	Features	`compute()` not `None`	`feature_window` (20)	`⏸ Feature compute returned None`
9	Volatile skip	regime not VOLATILE (if enabled)	`skip_volatile_regime` (False)	`⏸ Volatile regime — skipping`
10	Min confidence	`conf >= min_confidence`	`min_confidence` (0.63, `PO_MIN_CONF`)	`⏸ Low confidence: …`
11	Max confidence	`conf <= max_confidence`	`max_confidence` (0.85, `PO_MAX_CONF`)	`⏸ Overconfident: …`
12	Candle range	`high − low >= 0.00005`	none (hard-coded)	`⏸ Low range candle (…) — skip`
13	Indicator alignment	≥2 of 3 indicators agree	`require_indicator_alignment` (True, `PO_REQUIRE_ALIGNMENT`)	`⏸ Indicators misaligned (conf=…)`
14	Signal confirmation	N consecutive candles agree	`signal_confirmations` (1)	`⏸ Signal confirmation (k/N) …`
15	Consec-loss trigger	streak < `max_consec_losses`	`max_consec_losses` (3), `cooldown_seconds` (300 s)	`Hit N consecutive losses → cooldown`
16	5-min trend	SMA5 vs SMA20 of 5-min samples agrees with direction	none (hard-coded, needs ≥100 candles)	`⏸ 5-min trend DOWN vs CALL — skip`
17	Adaptive strategy	`should_trade()` returns true	learned (base: `min_confidence`)	`🧠 Adaptive skip: …`

Tuning notes

The static, money-related brakes (gates 2–6, 15) are your primary risk controls — see Risk management. Risk note: loosening them increases the money at risk per day; binary options losses are the full stake.
The signal-quality gates (10–14, 16) trade frequency for selectivity. Raising min_confidence or signal_confirmations produces fewer, stronger signals.
The learned gates (7, 17) tighten automatically after losses and relax after wins — see Adaptive strategy.

Signal Gates

The pipeline, in code order

Summary table

Tuning notes

See also