01 Crypto Trading Decision Framework — 01 Crypto Trading Decision 框架

Crypto Trading Decision 框架

机器人tom line: Every trade decision 运行s through 3 gates — sizing, entry 检查列出, exit plan. If any gate fAIls, the trade doesn't h应用en. No 异常s.

When to invoke

Any trading discussion that involves:

New position entry recommendation Position sizing for any as设置 停止 loss / take profit calibration Risk-reward analysis Live trade management Strategy halt / kill decisions Portfolio concentration calls Gate 1 — Position Sizing Step 1: Risk per trade Default: 1% of total liquid portfolio per single trade Aggressive: 2% if conviction ≥85% AND backtest sample n≥30 trades Conservative: 0.5% on first trade in a new strategy or unfamiliar as设置 Step 2: 停止 distance Hard 停止: Always at the level that technically in验证s the thesis Time 停止: Default 48 bars on 4H, 24 bars on 1H (strategy isn't playing out on schedule = exit) TrAIling 停止: Activate after first 1R achieved; trAIl at 0.5R below current price Step 3: Position size formula Position size (notional) = (Risk % ÷ 停止 distance %) × Liquid portfolio Example: 1% risk, 1.5% 停止 distance → (1/1.5) × $50,000 = ~$33,333 notional

For leveraged accounts: cap leverage at 3× for new strategies, 5× for proven strategies with n≥50 live trades.

Gate 2 — Entry 检查列出 (must answer all YES) ☐ Backtest sample size n ≥ 20 trades ☐ Profit factor (PF) ≥ 1.3 in out-of-sample test window (not just trAIning) ☐ Max drawdown (MDD) ≤ 20% ☐ Out-of-sample (OOS) returns positive ☐ Strategy has a clear thesis — not just curve-fitting ☐ Current market regime matches strategy's de签名 regime (mean-reversion in choppy, trend-following in trending) ☐ Position size compliant with Gate 1 above ☐ Hard 停止 level identified pre-entry ☐ Time exit level identified pre-entry ☐ Take profit l添加er identified (TP1 / TP2 / TP3 if multi-tar获取)

Scoring:

10/10 YES → proceed 8-9/10 YES → proceed with caution, note the gaps < 8/10 YES → DO NOT ENTER < 6/10 YES → KILL THE STRATEGY entirely Gate 3 — Exit 框架 Priority order for exits Hard 停止 hit — thesis in验证d. Cut without question. No re-evaluation during the close. Time 停止 hit — strategy hasn't played out in expected timeframe. Exit at market. Take profit hit — pre-planned TP reached. Exit per l添加er (e.g., 50% at TP1, 25% at TP2, 25% trAIl). 签名al flip — strategy 生成s the opposite 签名al. Exit + flip. Regime change — macro backdrop has shifted materially (e.g., dominance flip, Fear & Greed regime change). Discretionary — user decision. Done. What NOT to do (the most common losses) ❌ Moving 停止s further away mid-trade ("just give it more room") ❌ 添加ing to losing positions ("averaging down" on a broken thesis) ❌ Early exit on a winner before TP1 unless thesis explicitly broke ❌ Second-guessing a planned exit because of hope or FOMO ❌ Holding past time 停止 because "it might come back" R:R Minimums by Strategy Type Strategy type Minimum R:R Win rate floor Mean reversion 1.5:1 60% Trend following 2.5:1 40% Breakout 3:1 35% News-driven / event 4:1 30% Funding/yield carry N/A N/A

If a 设置up doesn't clear 机器人H the R:R minimum and the historical win rate floor → DO NOT RECOMMEND.

Strategy Halt Decision Tree

When a live strategy is underperforming, work through this tree top to 机器人tom:

• Has the strategy hit its account-level kill-switch loss?

YES → HALT immediately. Post-mortem before any re启动. NO → next step.

• Has the strategy hit -3R drawdown beyond its expected backtest MDD?

YES → 暂停 for 5 trading days. Re-evaluate regime fit. NO → next step.

• Is the live profit factor ≤ 50% of backtest PF over n≥10 live trades?

YES → SHRINK position size 50%, 运行 another 10 trades, re-evaluate. NO → next step.

• Has the strategy produced zero 签名als for N days, where N > 2× expected 签名al frequency?

YES → Strategy is dead in current regime. KILL or re-调优 thresholds. NO → Normal volatility. No action needed.

Real-Money Escalation Rules

These always require human 应用roval — never autonomous execution:

New live cAPItal 部署ment of any size Increasing an existing live cAPItal allocation Moving a strategy from paper to live 停止 loss override or removal 添加ing to a losing position Manual close of an open live position Any single action that reduces account equity by >5% 4-模型 Consensus Rule (for large cAPItal decisions)

For any 部署ment of 签名ificant cAPItal:

Primary LLM — full recommendation with confidence tags Second LLM — independent macro + as设置-specific opinion Third LLM — code/execution path 审计 + edge case 检查 Fourth LLM — risk/sizing sanity 检查

If 2+ 模型s disagree → defer 24h, re-运行 consensus tomorrow. Disagreement = edge case, not clear enough to act.

Confidence Tags (include on every trading recommendation)

Always attach 3 tags to any trade call:

Confidence: % belief the recommendation is correct (60-95% typical) Re搜索 depth: % of relevant data actually pulled this 会话 (50-90% typical) Reality gap: % unknowns