AxonTrade
An autonomous trading platform under ongoing rigorous testing and experimentation to prove robustness before it is trusted with capital: a validated strategy portfolio, a human-in-the-loop mode ladder, and a self-grading scorecard. Every decision is journaled, replayed against what actually happened, and scored against its own backtest reference.
The trading maths contains no language model
A deterministic rules engine trades daily bars on real market data through a simulated broker. LLMs sit strictly at the edges — explanations on explicit request, and one daily digest — routed through a single gateway. The engine's decisions never depend on them.
Deterministic decision engine
A portfolio of strategy/symbol cells — trend-following EMA, Donchian breakout, Bollinger mean-reversion and RSI mean-reversion — across FX, equities, an index and a commodity future, all on daily bars. Selected by nested walk-forward validation.
A risk engine that can only say no
It can veto or shrink a trade; it can never create one. Drawdown kill-switch, daily loss cap, position limits, risk-per-trade clamps, notional caps, and blackout windows that block entries only — exits are never blocked.
A mode ladder
SHADOW → HITL → AUTONOMOUS. Autonomy is earned in steps, with a human in the loop until the evidence says otherwise.
A self-grading scorecard
Every decision is replayed 20 bars later under worst-case bar ordering and compared against the strategy's own backtest profit factor. The system reports on itself.
A real session from the terminal
These figures are from an actual AxonTrade paper-trading run — reproduced here, not a live connection. Positions, risk limits, decision reasoning and the self-grading scorecard are exactly what the terminal shows. The strategy logic itself ("The Math") stays private.
| Ticket | Symbol | Side | Volume | Entry | Mark | SL / TP | P&L |
|---|---|---|---|---|---|---|---|
| 200002 | AA | SELL | 8.47 | 49.44500 | 50.32503 | 66.96711 / — | -7.45 |
| 200001 | NVDA | BUY | 6.00 | 219.97200 | 217.52824 | 203.46664 / — | -14.66 |
| 200004 | ^DJI | BUY | 0.14 | 53367.17600 | 53554.63300 | 52522.46000 / — | +26.24 |
| 200008 | QQQ | BUY | 72.32 | 712.25120 | 716.35834 | 705.76996 / — | +297.03 |
| 200005 | USDJPY | SELL | 0.15 | 159.54100 | 160.04400 | 160.66900 / — | -47.14 |
| Time | Strategy | Symbol | Action | Regime | Entry | SL / TP | Mode |
|---|---|---|---|---|---|---|---|
| Aug 29, 04:43 AM | breakout-donchian | USDCHF | SELL | RANGING | 0.80376 | 0.81102 / — | AUTONOMOUS |
What the engine saw: On the USDCHF daily chart, the Donchian breakout cell registered a downside break — price pushing through the lower edge of its recent daily range at 0.80376. The engine tagged the broader regime as RANGING with a confluence score of 0, meaning no supporting signals lined up behind the breakout; the confidence figure of 1 reflects the rule firing cleanly, not a strong multi-factor case.
Why SELL: This is a pure rule-based short entry. The breakout policy triggers on the range break itself, so once price closed beyond the channel boundary the target was set to -1 (short) and the order was queued autonomously. Note the honest caveat: breakout rules are weakest in ranging conditions, where breaks often fail and pull back inside the range. Nothing here confirms a trending follow-through — the engine took the signal because the rule fired, not because the backdrop was ideal.
Risk plan: Stop-loss sits at 0.81102, about 0.00726 above entry (roughly 72–73 pips), with position size set at 0.14 lots for a risk amount of about 126.62. The risk module approved with no veto reasons and the kill switch is inactive. There is no take-profit or reward-to-risk figure defined for this trade, so the exit is managed by the stop and whatever the strategy's own exit logic dictates — worth watching, since an open-ended target on a low-confluence breakout in a range means the stop is doing all the protective work.
| Time | Source | Headline — AA first |
|---|---|---|
| Aug 28, 01:32 AM | Yahoo · AA | South32 H2 Earnings Call Highlights |
| Aug 27, 06:00 PM | Yahoo · AA | Alcoa to Participate in Jefferies Global Industrials Conference 2026 |
| Aug 27, 04:34 AM | Yahoo · AA | Agilent Technologies Q3 Earnings Call Highlights |
| Aug 27, 12:05 AM | Yahoo · AA | Alcoa's Gallium Project Opens a New Door Beyond Aluminum |
| Aug 26, 06:30 PM | Yahoo · AA | Alcoa Stock Price Prediction: Middle East Energy Outlook Pushes AA to $68 |
| Aug 25, 02:09 PM | Yahoo · AA | Alcoa starts construction on gallium plant at Wagerup refinery |
| Aug 25, 02:24 AM | Yahoo · AA | U.S. Steel Stocks Nucor, Steel Dynamics Give Up Initial Gains On Canada Trade Talks Collapse |
| Aug 24, 06:40 PM | Yahoo · AA | Alcoa (AA) Breaks Ground On Wagerup Gallium Plant And Expands Critical Minerals Reach |
| Aug 24, 03:18 PM | Yahoo · AA | Australia, Japan, the United States and Alcoa Break Ground on Gallium Project in Western Australia |
| Aug 23, 09:37 PM | Yahoo · AA | Alcoa (AA) Stock Still Trades At A Discount To Fair Value After A 90% Run |
| Aug 23, 08:38 PM | Yahoo · AA | Alcoa (AA) Jumped, But What Is Drawing Attention Now? |
Outcomes in progress — every entry is replayed with worst-case ordering; grades land on SL/TP touch or after 20 daily bars. Pending — 23
| Symbol / Cell | Direction | Entry | SL | Bars | Unrealized R | MFE / MAE | Grade by |
|---|---|---|---|---|---|---|---|
| USDCHF meanrev-bollinger|USDCHF|D1 | BUY | 0.80376 | 0.77790 | 0/20 | — | — / — | Sep 24 |
| USDCHF breakout-donchian|USDCHF|D1 | SELL | 0.80376 | 0.81102 | 0/20 | — | — / — | Sep 24 |
| USDCHF meanrev-bollinger|USDCHF|D1 | BUY | 0.80383 | 0.77790 | 1/20 | -0.00 | +0.19 / -0.02 | Sep 23 |
| USDCHF breakout-donchian|USDCHF|D1 | SELL | 0.80383 | 0.81102 | 1/20 | +0.01 | +0.09 / -0.67 | Sep 23 |
| USDCHF meanrev-bollinger|USDCHF|D1 | BUY | 0.80505 | 0.77790 | 2/20 | -0.05 | +0.13 / -0.08 | Sep 22 |
| USDCHF breakout-donchian|USDCHF|D1 | SELL | 0.80505 | 0.81102 | 2/20 | +0.22 | +0.34 / -0.60 | Sep 22 |
| USDCHF meanrev-bollinger|USDCHF|D1 | BUY | 0.80178 | 0.77790 | 3/20 | +0.08 | +0.29 / -0.02 | Sep 21 |
| QQQ trend-ema|QQQ|D1 | BUY | 711.220 | 705.770 | 3/20 | +1.81 | +1.86 / -0.69 | Sep 21 |
| USDCHF meanrev-bollinger|USDCHF|D1 | BUY | 0.80217 | 0.77790 | 4/20 | +0.07 | +0.27 / -0.03 | Sep 20 |
| USDCHF breakout-donchian|USDCHF|D1 | SELL | 0.80217 | 0.81102 | 4/20 | -0.18 | +0.09 / -0.73 | Sep 20 |
| USDJPY breakout-donchian|USDJPY|D1 | SELL | 159.529 | 160.669 | 8/20 | +0.18 | +1.30 / -0.46 | Sep 14 |
| ^DJI trend-ema|^DJI|D1 | BUY | 53,354.43 | 52,522.46 | 8/20 | +0.26 | +0.43 / -0.72 | Sep 14 |
Each entry risks 0.50% of equity, defined by the distance to its stop.
Enabled YES · 9 events tracked. No blackout window active.
Read-only view — limits are enforced by the risk engine and configured elsewhere.
What "rigorous" means: publishing the gate that failed
Pooled walk-forward validation produced a profit factor of 1.68 over 132 trades — but the cell-consistency gate failed at 47%. That is why the system is accumulating live paper evidence rather than trading real money. We publish the gate that failed, not only the number that passed.
A foundation model was evaluated for direction and volatility forecasting, including a fine-tune. It reached roughly a 51% hit rate on a decisive sample, its volatility forecasts lost to plain EWMA, and none of six recalibration configurations passed. It is not wired into trading — and the reports are kept as the evidence for why.