crypto-trading-decision-framework

# Crypto Trading Decision Framework **Bottom line:** Every trade decision runs through 3 gates — sizing, entry checklist, exit plan. If any gate fails, the trade doesn't happen. No exceptions. --- ## When to invoke Any trading discussion that involves: - New position entry recommendation - Position sizing for any asset - Stop 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 asset ### Step 2: Stop distance - **Hard stop:** Always at the level that technically invalidates the thesis - **Time stop:** Default 48 bars on 4H, 24 bars on 1H (strategy isn't playing out on schedule = exit) - **Trailing stop:** Activate after first 1R achieved; trail at 0.5R below current price ### Step 3: Position size formula ``` Position size (notional) = (Risk % ÷ Stop distance %) × Liquid portfolio Example: 1% risk, 1.5% stop 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 Checklist (must answer all YES) 1. ☐ Backtest sample size n ≥ 20 trades 2. ☐ Profit factor (PF) ≥ 1.3 in out-of-sample test window (not just training) 3. ☐ Max drawdown (MDD) ≤ 20% 4. ☐ Out-of-sample (OOS) returns positive 5. ☐ Strategy has a clear thesis — not just curve-fitting 6. ☐ Current market regime matches strategy's design regime (mean-reversion in choppy, trend-following in trending) 7. ☐ Position size compliant with Gate 1 above 8. ☐ Hard stop level identified pre-entry 9. ☐ Time exit level identified pre-entry 10. ☐ Take profit ladder identified (TP1 / TP2 / TP3 if multi-target) **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 Framework ### Priority order for exits 1. **Hard stop hit** — thesis invalidated. Cut without question. No re-evaluation during the close. 2. **Time stop hit** — strategy hasn't played out in expected timeframe. Exit at market. 3. **Take profit hit** — pre-planned TP reached. Exit per ladder (e.g., 50% at TP1, 25% at TP2, 25% trail). 4. **Signal flip** — strategy generates the opposite signal. Exit + flip. 5. **Regime change** — macro backdrop has shifted materially (e.g., dominance flip, Fear & Greed regime change). 6. **Discretionary** — user decision. Done. ### What NOT to do (the most common losses) - ❌ Moving stops further away mid-trade ("just give it more room") - ❌ Adding 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 stop 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 setup doesn't clear BOTH 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 bottom: ``` 1. Has the strategy hit its account-level kill-switch loss? YES → HALT immediately. Post-mortem before any restart. NO → next step. 2. Has the strategy hit -3R drawdown beyond its expected backtest MDD? YES → PAUSE for 5 trading days. Re-evaluate regime fit. NO → next step. 3. Is the live profit factor ≤ 50% of backtest PF over n≥10 live trades? YES → SHRINK position size 50%, run another 10 trades, re-evaluate. NO → next step. 4. Has the strategy produced zero signals for N days, where N > 2× expected signal frequency? YES → Strategy is dead in current regime. KILL or re-tune thresholds. NO → Normal volatility. No action needed. ``` --- ## Real-Money Escalation Rules These always require human approval — never autonomous execution: - New live capital deployment of any size - Increasing an existing live capital allocation - Moving a strategy from paper to live - Stop loss override or removal - Adding to a losing position - Manual close of an open live position - Any single action that reduces account equity by >5% --- ## 4-Model Consensus Rule (for large capital decisions) For any deployment of significant capital: 1. **Primary LLM** — full recommendation with confidence tags 2. **Second LLM** — independent macro + asset-specific opinion 3. **Third LLM** — code/execution path audit + edge case check 4. **Fourth LLM** — risk/sizing sanity check If 2+ models disagree → defer 24h, re-run 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) - **Research depth:** % of relevant data actually pulled this session (50-90% typical) - **Reality gap:** % unknowns / black-swan exposure (5-25% typical) Example: "Confidence 82% / Research Depth 75% / Reality Gap 20%" This keeps recommendations honest and prevents overconfidence drift. --- ## Output Format When this framework produces a trade recommendation, structure it as: ``` ## Trade Recommendation: [ASSET] [LONG/SHORT] **Thesis:** [1-2 sentences — why this setup exists] **Regime fit:** [why current market supports this strategy type] **Sizing:** - Portfolio size: $X - Risk per trade: $X (1%) - Stop distance: X% - Position notional: $X **Entry checklist:** X/10 YES [list any NO items] **R:R:** X:1 [minimum met: YES/NO] **Levels:** - Entry: $X - Hard stop: $X (thesis invalidated if price reaches here because: [reason]) - TP1: $X (partial exit X%) - TP2: $X (partial exit X%) - Time stop: [date/bar count] **Confidence:** X% / Research Depth X% / Reality Gap X% **Escalation required:** YES/NO [why] ```