All picks are queried from tournament_picks table (MySQL: ejaguiar1_stocks, 3,149 rows, 34 AI models, 9 asset classes). Only forward-test OPEN picks are used. Synthetic/backtest data (SYNTHETIC_SEED_ENRICHED, BACKTEST_VERIFIED) and KILLED personas (ml_pattern, relative_strength, dividend_compound) are excluded. FOREX is blocked by kill gate (57.3% WR, -0.39% avg PnL).
| Method | Description | Picks |
|---|---|---|
| PERSONA WR | When persona has n≥20 resolved picks, confidence = persona win rate | 6 picks (PG 64%, SOL 65%, TLT/SPY/GLD/SHY 62.5%) |
| MODEL-REPORTED | Pick comes from model that reported its own confidence (HIGH/MEDIUM/LOW string or 0-1 float) | 7 picks (MSFT 30%, XOM 30%, CL=F 80%) |
| IMPUTED | No WR and no model-reported confidence — coin-flip assumption | 9 picks (SI=F 50%, PENNY/FUTURES 0%) |
Composite Score = Confidence × WR × RR × ln(n+1), normalized per asset class. Picks with n=0 score zero by definition.
Round 1 (7 models): Risk Manager + Portfolio Manager + Cerebras GPT-OSS-120B + DeepSeek V4 + KiloCode + Kimi + Cursor — debated which picks are safest, which to veto. Produced consensus top 5 and systemic issue list.
Round 2 (2 models): Multi-Asset Allocator + Financial Data Architect — expanded to all 9 asset classes, identified IPO/mutual fund infrastructure gaps.
Round 3 (3 models): Quant Researcher (EV/Sharpe/Kelly) + Behavioral Analyst (market narrative) + Hedge Fund PM ($500k AUM allocation). Produced 8-position risk-parity portfolio.
Round 4: IPO Lockup Expiry Strategy — SHORT 30 days before 180-day lockup expiry. Currently data-starved (needs live SEC EDGAR scraper).
Rounds 5-14 (10 agents): Pick-by-pick review, cross-round pattern analysis, devil's advocate audit, entry criteria standardization, data gap ranking, statistical edge recalculation, model attribution scorecard, final executive synthesis.
| Rank | Pick | Direction | Entry | WR | n | RR | Conf | Status |
|---|---|---|---|---|---|---|---|---|
| 1 | META | LONG | $620.71 | 60% | 124 | 1.7 | 50% | PROVEN |
| 2 | PG | SHORT | $167.37 | 64% | 164 | 1.5 | 64% | VERIFIED |
| 3 | GOOGL | LONG | $186.63 | 60% | 124 | 2.5 | 30% | ESTIMATED |
Best equity edge: PG SHORT — only verified-WR equity pick (64%, n=164). META LONG is the AI monetization flywheel play. GOOGL has highest RR (2.5) but low confidence (30%). Gap: UEPS fundamental screen shows ADBE (Score 0.839) as theoretically highest-quality but has 0 forward-test data in tournament_picks.
| Rank | Pick | Direction | Entry | WR | n | RR | Conf | Status |
|---|---|---|---|---|---|---|---|---|
| 1 | SOLUSDT | LONG | $157.39 | 65% | 23 | 2.1 | 65% | VERIFIED |
| 2 | AVAXUSDT | SHORT | $22.83 | 65% | 23 | 1.8 | 65% | SMALL N |
SOLUSDT LONG is the only crypto pick with verified WR (65%, n=23). vol_arb persona has only 23 resolved picks — statistically insufficient (95% CI: ±20%). Risk: Crypto shorts (AVAX, BTC, ETH) conflict directionally with SOL long — if risk-on returns, all shorts get run over.
💡 ELI5: SPY is the whole US stock market. Shorting it means you think stocks will go DOWN. GLD is gold — people buy it when they're scared about inflation or war. Together they're betting that stocks fall and gold rises, which is the "stagflation" playbook.
| Rank | Pick | Direction | Entry | WR | n | RR | Conf |
|---|---|---|---|---|---|---|---|
| 1 | SPY | SHORT | $726.80 | 62.5% | 124 | 1.9 | 62.5% |
| 2 | GLD | LONG | $257.93 | 62.5% | 124 | 1.2 | 62.5% |
SPY SHORT + GLD LONG = textbook stagflation pair. Both anchored by risk_parity persona (n=124). SPY SHORT is the highest composite score in the entire book. GLD has weakest RR (1.2) but diversification value.
| Rank | Pick | Direction | Entry | WR | n | RR | Conf |
|---|---|---|---|---|---|---|---|
| 1 | TLT | LONG | $87.66 | 62.5% | 124 | 1.8 | 62.5% |
| 2 | SHY | SHORT | $82.36 | 62.5% | 124 | 1.6 | 62.5% |
TLT LONG is the strongest consensus pick (7/7 models). SHY SHORT + TLT LONG = curve steepener. Risk: Bond Sharpe ratios inflated by low vol assumption (0.5% daily). Real bond strategies do not sustain Sharpe 19.
| Rank | Pick | Direction | Entry | WR | n | RR | Conf |
|---|---|---|---|---|---|---|---|
| 1 | SI=F | SHORT | $34.93 | 62.5% | 124 | 1.4 | 50% |
| 2 | CL=F | SHORT | $68.25 | 65% | 23 | 1.1 | 80% |
CL=F SHORT has near-zero EV (0.05 risk units) and RR=1.1 — risk/reward barely above breakeven. Gap: COMMODITY pipeline may be broken — top systems (multi_asset_cot, PF=4.72) have n=0 in resolved DB. All commodity picks are SHORT — check for regime bias. SI=F SHORT replaced CL=F in the adjusted portfolio.
WR=0%, n=0 for all 6 picks. MVST, KULR, QBTS (penny stocks) and ES=F, GC=F, CL=F (futures) have zero resolved data. No statistical basis for inclusion. These should be removed from active pick list until n≥50 resolved trades. Currently listed as PAPER ONLY — no real capital allocation.
Kill gate active: 57.3% WR, -0.39% avg PnL, 253 resolved picks. Statistical trap confirmed — many small wins, occasional large losers (3.2:1 loss-to-win ratio). 63% of FOREX wins are 1-basis-point "resolver flicker." Zero allocation until asymmetric TP/SL fix is validated.
| # | Issue | Severity | Fix |
|---|---|---|---|
| 1 | ML confidence INVERTED — 0.85-0.90 band has 20% WR | CRITICAL | Flip scoring: high confidence → sell signal. Recalibrate against realized outcomes. |
| 2 | No n-threshold gate — 0-data personas generating live signals | CRITICAL | Require ≥50 resolved trades per source before any signal passes. |
| 3 | PENNY/FUTURES with 0 resolved data | CRITICAL | Drop from active pick list. Paper-track until n≥50. |
| 4 | CL=F duplicated at 2 prices ($68.25 / $73) | HIGH | Keep commodity entry, drop futures entry. |
| 5 | FOREX resolver bug — 63% wins are 1bp "flicker" | HIGH | Already blocked by kill gate. Fix asymmetric TP/SL before re-enabling. |
| 6 | COMMODITY pipeline broken — top systems have n=0 in resolved DB | HIGH | Investigate table mismatch. multi_asset_cot PF=4.72 invisible to OOS validator. |
Resolver fires May 30 at 23:00 UTC.
| Pick | Direction | Prediction | Confidence | Rationale |
|---|---|---|---|---|
| TLT LONG | BULLISH | WIN | HIGH | 7/7 model consensus. Cleanest macro expression. |
| SPY SHORT | BEARISH | WIN | HIGH | Foundation of risk-off book. Market correction continuing. |
| GLD LONG | BULLISH | WIN | HIGH | Stagflation hedge thesis intact. |
| PG SHORT | BEARISH | WIN | MEDIUM | Only verified-WR equity pick (64%). |
| CL=F SHORT | BEARISH | LOSS | LOW | Near-zero EV. Oil geopolitical risk. |
| SHY SHORT | BEARISH | LOSS | LOW | Coin flip. Rate cut priced in. |
| File | Content |
|---|---|
reports/AI_HEDGE_FUND_SIMULATION_EXECUTIVE_SUMMARY_2026-05-24.md | Complete 159-line executive summary |
audit_dashboard/hedge_fund_simulation_20260524.html | 3-round debate results (7 models, per-agent insights) |
audit_dashboard/curated_picks_20260524.html | Top 3 picks per asset class |
updates/2026-05-24-cross-asset-statistical-analysis.md | Per-pick EV/Sharpe/Kelly, correlation matrix |
reports/CONFIDENCE_METHODOLOGY_2026-05-24.md | 3 confidence methods, thresholds, calibration gaps |
⚠️ Not financial advice. Educational/research simulation only. Zero real money deployed. AI Tournament · Curated Picks · Updates