---
name: trading-wisdom
description: Core trading insights learned from Agent Arena competition. Use when making any trading decision to apply institutional knowledge.
---

# Trading Wisdom

> Last updated: 2026-03-09 20:08 UTC
> Active patterns: 232
> Total samples: 22482
> Confidence threshold: 60%

## Key Learnings

1. Market was uniformly bullish (BNB +3.31%, BTC +2.68%, ETH +3.90%, SOL +5.07%, DOGE +2.46%) — the 7th consecutive window where regime misidentification was the primary loss driver.
2. Only 1 of 6 active agents was profitable (journal_aware +$55.48). The other 5 active agents lost a combined -$571.18, suggesting widespread SHORT bias despite uniformly positive market.
3. gptoss_skill_aware was the worst performer (-$340.51 on 31 trades), consistent with the persistent pattern of sophisticated validation frameworks providing false confidence on wrong-direction trades.
4. Self-reflective position management (journal_aware) continues to be the most reliable edge among active agents, now profitable in multiple bullish windows.
5. Zero trading (ta_bot, index_fund) outperformed 5 of 6 active agents, reinforcing that inaction beats wrong-direction action.
6. SOL was the best performer (+5.07%) — highest-beta assets continue to show the largest moves, making them both the best LONG targets and the worst SHORT targets.
7. Trade frequency amplifies losses when directional bias is wrong: skill_aware's 31 trades at -$10.98/trade vs contrarian's 6 trades at -$3.79/trade.

## Winning Strategies

### skill_aware_oss: High-frequency SHORT-biased tradi...
- **Confidence**: 95%
- **Total samples**: 200
- **Times confirmed**: 1
- **First seen**: 2026-02-01
- **Details**: skill_aware_oss: High-frequency SHORT-biased trading (200 trades/24h) with multi-timeframe bearish alignment validation, 2% equity risk sizing, and disciplined profit-taking. Achieved +$2911.52 in uniformly bearish market (-6% to -11% across all assets).

### Combining technical validation ('risk calculator s...
- **Confidence**: 92%
- **Total samples**: 200
- **Times confirmed**: 1
- **First seen**: 2026-02-01
- **Details**: Combining technical validation ('risk calculator shows 2:1 reward', 'validation permits trade') with trend alignment across 15m/1h/4h timeframes for SHORT entries in bearish markets. skill_aware_oss used this consistently with high confidence (0.78-0.92).

### index_fund's zero-trade strategy preserves capital...
- **Confidence**: 90%
- **Total samples**: 244
- **Times confirmed**: 1
- **First seen**: 2026-02-02
- **Details**: index_fund's zero-trade strategy preserves capital ($0 PnL) when all active trading agents lose money. In uniform_bullish markets (+1.33% to +3.82% across all assets), passive allocation beats active SHORT-biased trading.

### Position sizing at 2% equity risk with 2:1 reward ...
- **Confidence**: 90%
- **Total samples**: 183
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: Position sizing at 2% equity risk with 2:1 reward ratio SUCCEEDS when combined with validation checks AND market direction alignment. skill_aware_oss reasoning shows 'excellent 2:1 risk-reward with 2% equity risk' paired with bearish bias confirmation.

### Closing SHORT positions 'to lock in gains' when pr...
- **Confidence**: 88%
- **Total samples**: 172
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: Closing SHORT positions 'to lock in gains' when profitable is correct behavior in bearish markets. skill_aware_oss: 'closing BNBUSDT locks in profit and reduces exposure' with 0.92 confidence.

### Closing losing LONG positions quickly ('close exis...
- **Confidence**: 88%
- **Total samples**: 50
- **Times confirmed**: 1
- **First seen**: 2026-02-01
- **Details**: Closing losing LONG positions quickly ('close existing SOL long to free margin before initiating ETH short') to pivot to trend-aligned SHORT positions. skill_aware_oss demonstrated this adaptive behavior.

### skill_aware_oss maintains disciplined SHORT-biased...
- **Confidence**: 87%
- **Total samples**: 183
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: skill_aware_oss maintains disciplined SHORT-biased strategy in uniformly bearish markets (-7.95% to -16.34% across all assets), using multi-timeframe bearish confirmation with selective position management. Closes positions to 'lock in profit and reduce exposure' rather than panic-cutting small losses.

### skill_aware_oss achieves highest PnL (+$409.48) wi...
- **Confidence**: 85%
- **Total samples**: 172
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: skill_aware_oss achieves highest PnL (+$409.48) with high trade frequency (172 trades/24h) using SHORT-biased strategy in mixed-bullish market. Key: multi-timeframe bearish alignment (15m, 1h, 4h), 2:1 risk/reward ratio, and disciplined profit-taking on shorts when validation confirms.

### ta_baseline's low-frequency technical analysis app...
- **Confidence**: 85%
- **Total samples**: 38
- **Times confirmed**: 1
- **First seen**: 2026-02-02
- **Details**: ta_baseline's low-frequency technical analysis approach (38 trades/24h) minimizes losses in bullish markets where SHORT-biased strategies fail. Selective trading with clear technical signals outperforms high-frequency approaches when market direction contradicts agent bias.

### Zero trading preserves capital perfectly when acti...
- **Confidence**: 85%
- **Total samples**: 2
- **Times confirmed**: 1
- **First seen**: 2026-03-09
- **Details**: Zero trading preserves capital perfectly when active agents lose money. ta_bot (0 trades, $0) and index_fund (0 trades, $0) avoided all losses while 5 of 6 active agents lost money totaling -$571.18.

### Closing SHORT positions 'to lock in gains' when pr...
- **Confidence**: 83%
- **Total samples**: 172
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: Closing SHORT positions 'to lock in gains' when profitable combined with re-entering on bearish confirmation signals. skill_aware_oss demonstrates this cycle repeatedly with BTC shorts.

### skill_aware_oss achieves highest PnL (+$691.44) wi...
- **Confidence**: 82%
- **Total samples**: 49
- **Times confirmed**: 1
- **First seen**: 2026-01-30
- **Details**: skill_aware_oss achieves highest PnL (+$691.44) with moderate-high trade frequency (49 trades/24h) using SHORT positions in a uniformly bearish market. Key: Correctly identifying market direction and trading WITH the trend, not against it.

### Forum-aware selective trading with moderate freque...
- **Confidence**: 82%
- **Total samples**: 24
- **Times confirmed**: 1
- **First seen**: 2026-03-08
- **Details**: Forum-aware selective trading with moderate frequency (24 trades from 89 decisions, 27% action rate) in a mixed/slightly-negative market. gptoss_forum_aware achieved +$232.17, the best performer by far, suggesting forum/social sentiment signals correctly filtered out low-conviction trades and identified profitable opportunities.

### Ultra-low frequency or zero trading preserves capi...
- **Confidence**: 80%
- **Total samples**: 89
- **Times confirmed**: 1
- **First seen**: 2026-03-08
- **Details**: Ultra-low frequency or zero trading preserves capital when market direction is mixed/uncertain. index_fund (0 trades, $0 PnL) avoided all losses in a market where the momentum and contrarian agents lost money.

### ta_baseline achieves positive PnL (+$133.04) with ...
- **Confidence**: 80%
- **Total samples**: 36
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: ta_baseline achieves positive PnL (+$133.04) with moderate trade frequency (36 trades/24h) in mixed-bullish market. Technical analysis baseline with selective trading outperforms aggressive LLM-based strategies.

### ta_baseline achieves positive PnL ($+20.54) with e...
- **Confidence**: 80%
- **Total samples**: 20
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: ta_baseline achieves positive PnL ($+20.54) with extremely low trade frequency (20 trades vs 183 for top performer) by using simple SMA crossover signals and taking profit at defined ROE targets (6.1% ROE). Avoids overtrading in volatile bearish conditions.

### Zero trading or passive allocation preserves capit...
- **Confidence**: 80%
- **Total samples**: 12
- **Times confirmed**: 1
- **First seen**: 2026-03-07
- **Details**: Zero trading or passive allocation preserves capital in bearish markets. index_fund ($0 PnL, 12 trades all passive allocations) and ta_bot ($0 PnL, 0 trades) avoided all losses while active agents with wrong bias lost $12-$108.

### In uniformly bearish markets (all assets -1.3% to ...
- **Confidence**: 78%
- **Total samples**: 97
- **Times confirmed**: 1
- **First seen**: 2026-01-30
- **Details**: In uniformly bearish markets (all assets -1.3% to -2.3%), SHORT-biased strategies with RSI oversold confirmation and bearish MACD outperform. skill_aware_oss reasoning: 'RSI ~28, price below SMA, bearish MACD, multi-timeframe downtrend' for shorts.

### Journal-aware self-reflective position management ...
- **Confidence**: 78%
- **Total samples**: 27
- **Times confirmed**: 1
- **First seen**: 2026-03-09
- **Details**: Journal-aware self-reflective position management with moderate trade frequency (27 trades) in a uniformly bullish market (+2.5% to +5.1%). gptoss_journal_aware achieved +$55.48, the only profitable active agent, likely by recognizing overtrading tendencies and closing positions to reduce exposure.

### Moderate trade frequency (48-63 trades/24h) outper...
- **Confidence**: 77%
- **Total samples**: 160
- **Times confirmed**: 1
- **First seen**: 2026-01-30
- **Details**: Moderate trade frequency (48-63 trades/24h) outperforms both extremes in bearish markets. Too few trades (4-7) miss opportunities; the sweet spot captures trend moves without overtrading.

### Forum-aware selective trading with moderate freque...
- **Confidence**: 77%
- **Total samples**: 31
- **Times confirmed**: 1
- **First seen**: 2026-03-07
- **Details**: Forum-aware selective trading with moderate frequency (31 trades) in a uniformly bearish market. gptoss_forum_aware achieved +$250.04, the best performer, by likely aligning SHORT entries with the actual bearish trend and using social/forum sentiment to filter high-conviction trades.

### gptoss_20b_simple: Moderate trade frequency (149 t...
- **Confidence**: 75%
- **Total samples**: 149
- **Times confirmed**: 1
- **First seen**: 2026-02-01
- **Details**: gptoss_20b_simple: Moderate trade frequency (149 trades) with selective position management, achieving small positive PnL (+$8.60) by avoiding counter-trend trades and cutting losers early.

### gptoss_20b_simple achieves positive PnL (+$140.71)...
- **Confidence**: 75%
- **Total samples**: 48
- **Times confirmed**: 1
- **First seen**: 2026-01-30
- **Details**: gptoss_20b_simple achieves positive PnL (+$140.71) with 48 trades using disciplined position closing: 'Close the small profitable short to lock in gains' - taking profits on winning shorts rather than holding.

### gpt_simple's extremely low trade frequency (10 tra...
- **Confidence**: 75%
- **Total samples**: 10
- **Times confirmed**: 1
- **First seen**: 2026-02-02
- **Details**: gpt_simple's extremely low trade frequency (10 trades/24h) with selective entries limits losses to -$140.76 despite wrong-way positioning. Quick loss-cutting on BTC short ('BTC is grinding higher against my short') demonstrates adaptive behavior.

### Conservative trading (7 trades/24h) limits losses ...
- **Confidence**: 75%
- **Total samples**: 7
- **Times confirmed**: 1
- **First seen**: 2026-01-31
- **Details**: Conservative trading (7 trades/24h) limits losses in uncertain markets. gpt_simple's minimal activity resulted in only -$8.44 loss vs much larger losses from aggressive traders.

### Zero trading preserves capital perfectly when mark...
- **Confidence**: 75%
- **Total samples**: 2
- **Times confirmed**: 1
- **First seen**: 2026-03-06
- **Details**: Zero trading preserves capital perfectly when market direction is uncertain or agent lacks conviction. gptoss_skill_aware (0 trades, $0) and qwen35_skill_aware (0 trades, $0) avoided all losses by staying out entirely.

### Agentic multi-step reasoning with moderate trade f...
- **Confidence**: 72%
- **Total samples**: 30
- **Times confirmed**: 1
- **First seen**: 2026-03-08
- **Details**: Agentic multi-step reasoning with moderate trade frequency (30 trades) achieved +$123.26 in a mixed market. gptoss_agentic's sophisticated analysis appears to have correctly navigated the mixed regime with selective entries.

### Low trade frequency with correct directional aware...
- **Confidence**: 72%
- **Total samples**: 27
- **Times confirmed**: 1
- **First seen**: 2026-03-09
- **Details**: Low trade frequency with correct directional awareness. gptoss_journal_aware's 27 trades (+$55.48) vs gptoss_skill_aware's 31 trades (-$340.51) shows that self-reflective management outperforms sophisticated validation frameworks when both trade at similar frequencies.

### Journal-aware self-reflective position management ...
- **Confidence**: 67%
- **Total samples**: 48
- **Times confirmed**: 1
- **First seen**: 2026-03-06
- **Details**: Journal-aware self-reflective position management with high trade frequency (48 trades) in a uniformly bearish market. gptoss_journal_aware achieved +$296.01, the best performer, suggesting meta-cognitive exits ('close to reduce exposure', 'we are overtrading') correctly captured short-side profits and limited drawdowns when the market moved -3% to -5%.

### Low trade frequency with self-reflective or skill-...
- **Confidence**: 65%
- **Total samples**: 60
- **Times confirmed**: 1
- **First seen**: 2026-03-08
- **Details**: Low trade frequency with self-reflective or skill-aware management preserves capital and captures small gains. gptoss_journal_aware (29 trades, +$18.76) and gptoss_skill_aware (31 trades, +$22.43) both achieved modest positive PnL by keeping trade counts moderate.

### Agentic multi-step reasoning with moderate trade f...
- **Confidence**: 65%
- **Total samples**: 32
- **Times confirmed**: 1
- **First seen**: 2026-03-07
- **Details**: Agentic multi-step reasoning with moderate trade frequency (32 trades) achieved +$77.07 in a bearish market. gptoss_agentic's sophisticated analysis likely correctly identified bearish regime and managed positions with reasonable frequency.

### Agentic multi-step reasoning with moderate-high tr...
- **Confidence**: 63%
- **Total samples**: 30
- **Times confirmed**: 1
- **First seen**: 2026-03-06
- **Details**: Agentic multi-step reasoning with moderate-high trade frequency (30 trades) achieved +$193.68 in a uniformly bearish market. gptoss_agentic's sophisticated analysis likely correctly identified the bearish regime and executed SHORT-biased trades with reasonable frequency.

### Zero trading preserves capital perfectly when mark...
- **Confidence**: 60%
- **Total samples**: 2
- **Times confirmed**: 1
- **First seen**: 2026-03-05
- **Details**: Zero trading preserves capital perfectly when market is uniformly bearish and agent directional bias is uncertain. index_fund (0 trades, $0 PnL) and qwen35_skill_aware (1 trade, $0 PnL) avoided all losses while active agents lost $5.98 to $229.37.

### Journal-aware self-reflective position management ...
- **Confidence**: 55%
- **Total samples**: 32
- **Times confirmed**: 1
- **First seen**: 2026-03-07
- **Details**: Journal-aware self-reflective position management with 32 trades achieved +$1.77 — near breakeven but capital preservation in a bearish market where others lost significantly. Meta-cognitive exits limited damage.

### Forum-aware selective trading with moderate freque...
- **Confidence**: 55%
- **Total samples**: 26
- **Times confirmed**: 1
- **First seen**: 2026-03-06
- **Details**: Forum-aware selective trading with moderate frequency (26 trades) achieved +$31.87 in a bearish market. gptoss_forum_aware's social/forum sentiment signals likely provided correct bearish conviction without excessive overtrading.

### Ultra-low trade frequency with simple strategies i...
- **Confidence**: 52%
- **Total samples**: 29
- **Times confirmed**: 1
- **First seen**: 2026-03-05
- **Details**: Ultra-low trade frequency with simple strategies in a uniformly bearish market. qwen35_122b_simple (16 trades, +$50.27) and qwen35_simple (13 trades, +$35.65) were the only profitable agents, suggesting selective entries with correct directional alignment (likely SHORT-biased or quick profit-taking) outperform in moderate bearish conditions.

### Index fund passive allocation with minimal trading...
- **Confidence**: 50%
- **Total samples**: 8
- **Times confirmed**: 1
- **First seen**: 2026-01-28
- **Details**: Index fund passive allocation with minimal trading (8 trades/24h) achieves break-even ($0.00 PnL) in bullish markets where all assets are positive, outperforming all active strategies

### Low-frequency simple models (13-16 trades) with mo...
- **Confidence**: 48%
- **Total samples**: 29
- **Times confirmed**: 1
- **First seen**: 2026-03-05
- **Details**: Low-frequency simple models (13-16 trades) with modest positive PnL in a bearish market where all 5 assets declined -1.7% to -8.7%. These agents likely captured short-term moves or correctly timed SHORT entries without overtrading.

### Zero trading preserves capital when active agents ...
- **Confidence**: 42%
- **Total samples**: 8
- **Times confirmed**: 1
- **First seen**: 2026-03-04
- **Details**: Zero trading preserves capital when active agents lose money. ta_bot (0 trades, $0) and index_fund (8 trades, $0) avoided all losses while gptoss_agentic (-$435.19) and gptoss_forum_aware (-$403.29) suffered massive drawdowns.

## Patterns to Avoid

- **AVOID**: HIGH-CONFIDENCE SHORT entries during UNIFORM BULLISH market regime. skill_aware_oss opened shorts with 0.85-0.92 confidence on ETHUSDT/BTCUSDT citing 'strong bearish alignment' while market gained +1.33% to +3.82%. Result: -$963.36 (worst performer).
  - Conf: 95%, N=501, seen 1x
- **AVOID**: Validation checks passing ('risk calculator fits 2% equity', 'validation permits trade') used as primary trade justification. All losing agents cited validation success while making directionally wrong trades. Process compliance ≠ profitable trades.
  - Conf: 95%, N=366, seen 1x
- **AVOID**: High trade frequency (147-188 trades/24h) with SHORT bias in bullish markets compounds losses. Each trade against trend adds to drawdown. skill_aware_oss (166 trades, -$963) vs ta_baseline (38 trades, -$120) shows frequency amplifies directional error.
  - Conf: 93%, N=501, seen 1x
- **AVOID**: Opening LONG positions in uniformly bearish markets. agentic_gptoss opened SOL long with 0.95 confidence ('bullish bias with strong technical indicators') but market dropped -11.31%. Counter-trend trades destroy edge.
  - Conf: 92%, N=189, seen 1x
- **AVOID**: Multi-timeframe bearish alignment (15m, 1h, 4h) as SHORT entry signal FAILS in uniform_bullish markets. All three high-frequency agents cited this signal repeatedly while losing $700-$963. Short-term technical bearishness does not override macro bullish regime.
  - Conf: 92%, N=501, seen 1x
- **AVOID**: gpt_simple: Extremely low trade frequency (6 trades/24h) in strongly trending bearish market results in -$163.64 loss. Insufficient participation to capture directional moves.
  - Conf: 90%, N=6, seen 1x
- **AVOID**: Skill-aware agent with highest losses (-$340.51 on 31 trades, -$10.98/trade). Sophisticated risk frameworks (2% equity, 2:1 reward ratio) failed catastrophically, likely due to regime misidentification — citing 'uniform bearish' while market was uniformly bullish (+2.5% to +5.1%).
  - Conf: 90%, N=31, seen 1x
- **AVOID**: Cutting small losses quickly to 'free margin for future trades' destroys edge. gptoss_20b_simple repeatedly closes positions with reasoning like 'close the largest loss to c