如何批判性閱讀 AI 交易理由

AI gives a confident-sounding answer for every prompt. That's the hardest part: confidence is uncorrelated with correctness. You need a checklist before tapping BUY on any AI-generated trade.

The 4 questions

1. Does the rationale name a specific catalyst?

Strong: "MACD golden cross on 1H, price reclaiming EMA50 with rising volume on the breakout candle". Weak: "Momentum is improving and bulls are in control". The first describes what happened. The second describes a feeling.

2. Is the TP / SL geometry physically reasonable?

For XAU/USD the daily ATR is roughly $30-50. A 15-min scalp suggesting a $200 TP is a fantasy. Eyeball the recent candle range — your TP should fit inside 1-3 ATRs unless you're holding for days. GoldMind's R:R preview tells you in pips and dollars before you click; if the dollar reward looks crazy, the AI is dreaming.

3. Does the multi-timeframe resonance agree?

The 4-dot resonance widget at the top of the chart shows MACD state on 5m / 15m / 1h / 4h. If the AI says SHORT but 1h and 4h are clearly bullish (green dots), that's a counter-trend trade — it has a lower base-rate hit, demand a tighter stop and smaller size.

4. What's the news context?

Even a perfect technical setup gets steamrolled when the FOMC speaks. Glance at the News AI Ticker for your symbol — if there's a high-impact event in the next 4 hours, either skip the trade or cut size in half.

Worked example: XAU/USD long

AI says: "Long XAU/USD. 進場 2403, TP 2418, SL 2398. Reason: 4H MACD bullish cross + price holding above EMA20 + low DXY."

• Specific catalyst? Yes — names MACD, EMA20, DXY. ✓
• TP/SL geometry? $15 reward, $5 risk — R:R 3:1, well inside daily range. ✓
• Resonance? If 1H and 4H are green, agree. If 1H is red, downgrade to 0.5× size.
• News? If NFP is in 30 minutes, skip — slippage risk too high.

3 out of 4 agree → take it at planned size. 2 out of 4 → half size. 1 → don't take it. This simple checklist will eliminate most of the regret trades.

When AI is most useful

AI excels at summarizing the state of a chart you don't have time to look at — a 30-second answer to "is anything happening on USDJPY right now?". It's mediocre at predicting tomorrow's direction. Use it as a screener, not an oracle.