Cortex AI Strategy for NinjaTrader 8 — self-optimizing orders
v0.6.6 · 13 order-management actions · 5-Filter stack · Reinforcement Learning
Manual order management can negatively affect your trading results
You spot the setup. You enter. Then the stress begins: move the stop, set a partial exit, trail the runner, evaluate re-entry. Three manual decisions per trade — all under time pressure.
Stops moved too early or too late
Break-even lock on gut feeling. Trailing by hand. With a 4-tick drawdown you push the stop into profit — and get stopped out right there. With a runner, you keep the stop too wide and give back half the gain.
Setups without pre-entry filtering
Low ATR + Wyckoff range + POC pin — an experienced trader would skip all three. But in real time you cannot read that constellation. Result: trades in dead phases with negative expectancy.
No learning loop — every day starts at zero
You lose three times in a row with the same signal constellation. But tomorrow you take the same setup again. Without structured outcome analysis you repeat your patterns — the bad ones too.
Three protection layers between your capital and a bad trade
The Cortex AI Strategy is developed, tested and optimized exclusively for the ES (E-mini S&P 500). The stop/target distances and order-flow thresholds are tuned to the volatility, tick structure and liquidity of the ES — on other instruments such as the NQ the same settings can lead to instant stop-outs, so we make no claim of functionality there. The analysis tools do work on any instrument.
Cortex OEM AI intervenes in three places: before entry, during the trade, and after the trade. Every layer is measurable, visible in the NT8 chart, and configurable.
5-Filter Stack before every entry
Five parallel hard filters check every setup before an order is submitted. Trades in toxic market phases are blocked — not executed and rescued afterwards, but never entered in the first place.
- Low-ATR Filter (minimum volatility)
- Wyckoff Conflict (regime vs. phase)
- POC Pin Filter (±3 ticks from POC)
- Counter-Regime Defense (aiConf gate)
- Wyckoff Trap (SHORT+ACCUM, LONG+DIST)
13 order-management actions during the trade
The strategy adapts stop, targets, and trail logic live — based on Wyckoff regime, magnet position (DOM/iceberg), and trade maturity. A Counter-Flow Hard-Veto blocks magnet bypasses against a ≥±30% large-player bias; DOMINANT-WALL-SL anchors the stop to the dominant order-book wall (capped ≤24 ticks).
- STOP_TIGHTEN / STOP_LOOSEN
- MOVE_TO_BE / TRAIL_AGGRESSIVE / TRAIL_LOOSE
- T2_PULL_IN / T2_EXTEND
- QUICK_PROFIT / PARTIAL_EXIT
- PRE_STOP_EXIT / RE_ENTRY
- SKIP_SETUP / SUPPRESS_TRAIL
Own-model Reinforcement Learning after every trade
The RL engine learns exclusively from your own trades — from trade #1, fully active after 30 trades. No pretrained bundle. Signal-regime keys (e.g. Dlt_TREND_BEAR) get weights between 0.1 and 3.0.
- strategy_learn.json (user-persistent)
- NET SAVED lock-in credit
- Counterfactual V2 for PRE_STOP_EXIT
- Magnet Gate (DOM/iceberg bypass)
- RL active after 30 of your own trades
A real trade path — from setup to lock-in
Example trade from 2026-05-21, 17:36 CEST: LONG OFA_Normal during the Lunch phase. Four OEM actions fired in parallel to manage the position from entry to exit.
4 OEM actions in 9 minutes
The strategy detected the LUNCH phase + NEUTRAL regime and shrunk T1 to 12 ticks (instead of 14). At 9 ticks MFE the pre-T1 BE-Lock fired automatically and secured the stop at break-even. After T1 fill, STOP_TIGHTEN set the trail to +11 ticks lock-in.
- 17:36:04 · PHASE-T1-SHRINKLunch × NEUTRAL = 0.80 · T1 14T → 12T, BeTrigger 11T → 9T
- 17:38:22 · PRE-T1 BE-LOCKMFE = 9.0T = BeTrigger · Stop 7416 → 7418.50 (BE+0.5T)
- 17:41:55 · T1-FILLLimit @ 7422 filled · quantity halved, runner active
- 17:43:18 · OEM STOP_TIGHTENLock-in +11T · weight 1.44 → 1.56 (regime match)
- 17:45:09 · EXITRunner stop fired · OEM trail vs. default trail
What the code says, not what marketing claims
Every value in this table is directly verifiable in Strategy v0.6.6 — not from backtest marketing or estimated assumptions.
Common questions about Cortex AI Strategy
Most traders ask the same six questions before running the strategy live. Here are the answers.
How does the RiskLevel work and which levels exist?
There is a single master control: RiskLevel 1–3 (the internal RiskMode dropdown is hidden). Level 1 (Conservative) adds +2 to the minimum score and +1 to the required signal sources and makes the AI gate mandatory (HARD). Level 2 (Moderate) softens this to +1/+1 with a soft AI gate. Level 3 (Aggressive) applies no tightening.
The $-risk limits (MaxLoss, DayTarget, ProfitLock, Cooldown) are independent user inputs and are never overridden by the RiskLevel. Recommendation for the first 2–3 trading weeks: Level 2. Once your own RL data is collected and you understand the logs, switch to Level 3.
Does the strategy learn from a pretrained bundle?
No. There is no pretrained bundle anymore. The RL engine is an own-model: it learns exclusively from your own trades — recording starts at trade #1, and the reinforcement learning becomes fully active after 30 of your own trades.
Your weights are written to a strategy_learn.json in your user folder. Existing files are never overwritten — your own RL data is safe. In TrainingMode you can rehearse the logic purely in SIM (no real risk).
How exactly does the pre-T1 BE-Lock work?
As soon as the trade reaches an MFE (maximum favorable excursion) equal to the BeTrigger, the strategy automatically moves the stop to break-even +0.5 ticks. This happens before T1 fill — not afterwards.
In Lunch and Afternoon phases with NEUTRAL or RANGE regime, a matrix additionally shrinks the BeTrigger (e.g. Lunch × NEUTRAL = 0.80, so BeTrigger 11T → 9T). Result: smaller trades are protected earlier.
What is the Magnet Gate and why is it a bypass?
The indicator writes an indicator_magnet_state.json every 500 ms with DOM aggregation and iceberg position from the order book. The Magnet Gate reads this file: if a massive liquidity magnet (DOM cluster + iceberg) lies in trade direction above the entry AND your MFE is ≥ 8 ticks, the PRE_STOP_EXIT logic is deliberately skipped.
Reasoning: in this constellation the market wants to push higher. An early exit would undermine the reversal USP of the CORTEX engine.
Which instruments does the strategy run on?
Primarily tested and tradeable on ES (E-Mini S&P 500) and NQ (Nasdaq Futures). Session detection runs in NY time (see GetETTod()), display in your NT8 timezone.
Crude Oil (CL) and MES/MNQ work as well, but the default bracket values (SL/T1/T2 in ticks) are calibrated to the ES tick value. For smaller/larger tick values, adjust the RegimeBrackets dict defaults.
Which plan do I need and what does the strategy cost?
The Cortex AI Strategy is Pro and Institutional only — not in Trial. The Trial uses the Trial NT8 ZIP (Bridge + Indicator v0.3.13, without the auto-strategy) and is only license/time-limited.
Pro pricing: €99/month, €890/year (instead of €1,188) or €2,997 Founder Lifetime (1/5 sold). All prices exclude 21 % IVA. Pro/Trial support +€49/month per additional PC (1 PC per license).