In Plain English: What We Found (a 90-second read for anyone, no finance background needed)
Imagine you manage a $1 billion fund. Your job: beat the S&P 500 (buy & hold = "do nothing" baseline).
We investigated 9 different investment worlds (crypto, stocks, bonds, commodities, currencies, ETFs, futures, memecoins). Here's the honest answer:
📋 Show the plain-English summary
- CRYPTO SHORT (our strongest signal): During crypto crashes, coins briefly bounce (like a car bouncing once before hitting the wall). If we short at that bounce (RSI 30-50), it makes +93.5% in 3 months. BUT — the worst loss was 49%. Too scary for big money.
- STOCK BREAKOUTS: Buying stocks that break out of consolidation works (55% win rate, 2x profit factor), but we only have 58 trades. Need 100+ before trusting it.
- ETF TACTICAL ROTATION: Not stock-picking — it's a "move your money to safety" system. Beats S&P 500 on risk-adjusted returns (Sharpe 1.48 vs 1.37). Ready to ship as a product.
- 13F INSTITUTIONAL FLOW: Following where big mutual funds put money (INTC, XOM). Ready to productize as a fund screener.
- EVERYTHING ELSE: Commodities, forex, bonds, futures, memecoins — lost money or had fake results that don't count.
Bottom line: 0/9 investment worlds have a PROVEN, ready-to-trade strategy. We have one signal with real edge (crypto SHORT), but it's too risky for large capital without fixes. We have two shippable products (ETF Tactical Rotation + 13F Fund Ranker). No strategy is "money-ready" yet.
❓ Having trouble with the jargon? Quick glossary (hover or expand to learn)
| PF (Profit Factor) | How much you make divided by how much you lose. PF=2 means you make $2 for every $1 you lose. PF must be >1.0 to be profitable. | |
| WR (Win Rate) | What % of trades are winners. 60% WR means 60 out of 100 trades end in profit. | |
| MaxDD (Max Drawdown) | The worst peak-to-trough drop. 49% MaxDD means at one point your money was almost cut in half. | |
| CI-LB (CI Lower Bound) | If we replayed history 10,000 times, the lowest PF we'd see 95% of the time. Must be above 1.0 to say edge is "real." | |
| DSR (Deflated Sharpe Ratio) | Sharpe ratio after accounting for testing 2,000+ strategies. If 1,999 are fake, one will look good by luck alone — DSR corrects for that. | |
| RSI (Relative Strength Index) | A "momentum speedometer" from 0-100. Below 30 = extremely oversold (price crashed hard). 30-50 = modestly oversold. 50-70 = neutral. | |
| SHORT position | Betting a price goes DOWN. You profit when the asset falls. Opposite of "LONG" (betting it goes up). | |
| 13F filing | Quarterly disclosure that hedge funds file showing what stocks they bought/sold. We track new institutional buying as a signal. |
What the labels mean (tier definitions — what each verdict means in plain English)
🟢 PROVEN
Forward-confirmed: passes ALL gates, forward cohort matured (n≥100+ outcomes), no decay. Safe to consider for small live sizing.
🟡 CANDIDATE
Backtest looks strong but fails ≥1 gate (drawdown, costs, n too small, or no forward proof). Needs more data.
🔴 FAIL / REJECTED
Fails data integrity or statistical gates. Artifact, coin-flip, or refuted by peers. Do NOT trade.
⚫ COIN-FLIP
WR ~50%, PF ~1.0, no edge. Not enough evidence.
🔵 RESEARCH-VALIDATED
Strong retrospective signal with economic rationale; forward cohort running. Product candidate.
⚪ NOT_READY
Too early — needs more forward outcomes or data quality fix.
Asset class snapshot (MySQL [REDACTED], intrabar-resolved, hard-only TP/SL)
| Asset class | Best candidate | n (hard) | WR | PF (net) | Direction | Label |
|---|---|---|---|---|---|---|
| CRYPTO | crypto_short_rsi3050 | 125 | 60.8% | 2.16 | SHORT | CANDIDATE |
| EQUITY | rs-breakout-scout (stock) | 58 | 55.2% | 2.00 | LONG | CANDIDATE (weak) |
| ETF (Tactical) | MRBB+200d TAA | 487 | — | 1.84 | LONG | RESEARCH-VAL |
| COMMODITY | commodity_rsi_divergence | 16 | 56.2% | 3.24 | LONG | CANDIDATE (weak) |
| FOREX | forex_trend_aligned | 50 | 60.0% | 2.32 | LONG | CANDIDATE (weak) |
| BOND | — | 27 | 22.2% | 0.85 | — | FAIL |
| FUTURES | — | 90 | 58.9% | 0.37 | — | FAIL |
| MEMECOIN | — | 48 | 54.2% | 0.44 | — | FAIL |
| ETF (directional) | — | 27 | 14.8% | 0.59 | — | FAIL |
1. CRYPTO — crypto_short_rsi3050 (CANDIDATE, fails MaxDD)
Direction: SHORT · Entry: RSI(14,1h) enters 30-50 oversold band · Symbols: 31 (DOT, WIF, ADA, STX, DOGE, etc.)
crypto_short_rsi3050
CANDIDATE (weak)
net PF 2.16 (vs 1.68 baseline SHORT)
Entry condition: CRYPTO SHORT when RSI(14,1h) is in the 30-50 oversold band. Found by
INCEPTION/ZOOCODE mining (2026-10-22). The sibling RSI 50-70 SHORT has
net PF 0.95 — it only wins in downtrends. RSI 30-50 SHORT wins in BOTH regimes.
2.16
net PF (Sharpe 6.9, Sortino 18.3)
n (trades)
125
125
Win Rate
60.8% (76 TP / 49 SL)
60.8% (76 TP / 49 SL)
Avg win/loss
2.45% / 1.48%
2.45% / 1.48%
Expectancy
+0.75%/trade
+0.75%/trade
Max DD
49.0%
49.0%
Symbols
31
31
Conc (top-3)
39.4%
39.4%
CI-LB (bootstrap)
1.43
1.43
L30d net PF
4.97
4.97
Direction
SHORT
SHORT
Time active
93%
93%
Total return
+93.5%
+93.5%
CVaR (worst 5%)
-4.97%
-4.97%
🔧 Technical reasoning — WHY RSI 30-50 SHORT beats everything else
| Component | Plain-language explanation | ELI5 |
|---|---|---|
| RSI(14,1h) = 30-50 | RSI measures "how oversold is this coin?" On 1-hour bars, RSI below 30 = extremely oversold, 30-50 = modestly oversold, 50-70 = neutral, 70+ = overbought. Going SHORT when RSI is already in the 30-50 (oversold) zone sounds counterintuitive — but it captures the moment when a brief relief bounce stalls and the downtrend resumes. The RSI 50-70 SHORT condition only works during established downtrends; RSI 30-50 SHORT works because the next 1h candle often confirms the bounce failed. |
Like selling a stock that just bounced 5% off its lows — it often drops back as sellers return. |
| Direction = SHORT | In crypto bear markets (May-Jul 2026), altcoins get washed out to RSI 30-40 and then "dead cat bounce" to 45-50 before continuing down. SHORT captures that second leg. |
Shorting = betting the price goes down. When a coin bounces off a deep discount, it often keeps going down. |
| Why NOT RSI 50-70 SHORT? (the refuted sibling) | RSI 50-70 SHORT has net PF 0.951 — it LOSES money. It only works when BTC is in a downtrend (wins in May-Jul, loses in Aug-Oct). RSI 30-50 SHORT wins in BOTH regimes (it's a mean-reversion signal, not a trend-following one). |
The 50-70 range SHORT is regime-confined. The 30-50 SHORT is regime-agnostic. |
| momentum-WITH variant (F2 WITH) | Adding "price momentum confirms the SHORT direction" boosts net PF to 2.475 (n=102). STRONGER signal, fewer samples. |
Only short when momentum agrees the trend is down. Filter within a filter. |
💰 Fundamental / economic rationale
Why this edge exists (economic story):
- Leverage-cascade liquidations: When altcoins dump hard (RSI → 30), leveraged longs get liquidated. The brief relief bounce to RSI ~45-50 triggers secondary liquidations → cascading SHORT opportunity.
- Funding rate dynamics: During dumps, perps funding turns deeply negative. Retail shorts pay positive funding but the cascade continues. Entry at RSI 30-50 catches the next wave.
- Liquidity evaporation: Low-liquidity altcoins gap down hard. SHORT entered at RSI 30-50 (not at the bottom) avoids the "catching a falling knife" risk.
- Contrarian vs trend-following: Most crypto strategies are LONG (trend-following). Going SHORT in oversold conditions is contrarian to the dominant paradigm → less crowded, more reliable.
⚖️ Grok's 13-gate institutional scorecard
| # | Gate / Metric | Result | ELI5 explanation |
|---|---|---|---|
| S1 | n ≥ 100 (≥500 to scale) | ✅ 125 | Need 100+ trades to separate skill from luck. 125 passes. |
| S2 | PF ≥ 1.5 net of costs | ✅ 2.16 | Make $2.16 gross for every $1 lost, after paying all trading fees. |
| S3 | PF CI lower bound ≥ 1.2 | ✅ 1.43 | If history repeats 10,000 times, PF still stays above 1.2 → real edge. |
| S4 | WR ≥ 50% | ✅ 60.8% | More than half the trades are winners. |
| S5 | Max Drawdown ≤ 20% | ❌ 49.0% | Portfolio dropped 49% at its worst — WAY too risky for a big fund. |
| S6 | Random-entry null ≥ 95th pct | ✅ | Beats randomly entering at the same time 95% of the time. |
| S7 | Permutation test p < 0.05 | ✅ | Shuffling trade labels randomly makes the edge disappear → it's real. |
| S8 | DSR > 0.95 | ✅ ~1.0 | After accounting for 2,000+ strategies tested, the signal still survives. |
| S10 | PF ≥ 1.3 at 2× costs | ✅ ~1.8 (est) | Still profitable if fees double (conservative). |
| S11 | Concentration: top ≤ 30% | ✅ 39.4% | Close to the 30% limit — 5 coins (DOT, WIF, ADA, STX, DOGE) carry 54% of P&L. |
| S12 | Stability: 70% months PF>1 | ❌ 4/6 months | 4 winning months, 2 losing — borderline. Fails the 70% stability gate. |
| S13 | Capacity / impact | ⚠️ Unmeasured | Small-cap altcoins can't absorb large capital. Capacity unknown. |
📉 Monthly time-series (vs BTC benchmark)
| Month | Strategy return | BTC regime | Verdict |
|---|---|---|---|
| 2026-05 | +1.08% | BTC ↓ | Win (small) |
| 2026-06 | +336.6% | BTC ↓↓ | Win (huge) |
| 2026-07 | +210.2% | BTC ↓↓ | Win (huge) |
| 2026-08 | -82.7% | BTC ↑ | Loss |
| 2026-09 | -131.0% | BTC ↑↑ | Loss (big) |
| 2026-10 (L30d) | +117.0% | BTC ↓ | Win (huge, recent) |
This strategy is a bear-market short. It wins massively when crypto crashes (Jun-Jul) and loses when crypto rallies (Aug-Sep). If deployed as a constant SHORT, it would need a regime filter to pause when BTC is rising.
📋 Rejected crypto alternatives (why we didn't pick them)
| Rejected strategy | Claimed | Why it fails |
|---|---|---|
| mega_mutation (Poolside original) | n=171, WR 80.7%, PF 7.66 | Grok R3 FAIL: created_at=NULL for ALL rows. Symbols chosen after the fact. Base strategy: n=24, WR 15.4%, PF 0.42. |
| high_trust_momentum (OpenCode PROVEN) | n=206, WR 58.7%, PF 1.81 | Pool inflates n by 79 TIME_EXIT soft-closes. Hard-only n=54. Oct decay: WR 38.1% (Buffy). CI crosses 0. |
| luxalgo_confluence SHORT | n=572, net PF 0.97 | Net PF < 1.0 after costs → loses money. Not viable. |
| momentum_breakout | n=113, WR 100%, PF 999 | 1 distinct day = frozen-exit artifact. Auto-reject. |
| crypto alpha_engine cell (conf .75-.80) | n=185, WR 75%, PF 3.90 | Pre-resolver-fix data. 100% WR on 9 close days. Artifact risk HIGH. |
2. EQUITY — rs-breakout-scout (CANDIDATE weak) + 13F fund-flow angle
Direction: LONG · Symbols: XOM, AAPL, others · Timeframe: daily swing (2-14 days)
rs-breakout-scout
CANDIDATE (weak)
PF 2.00, n=58
Found via hidden-winner scan (Grok's master plan, §10). Not shown on /audit's strategy table.
Top symbol XOM at 21% of PnL — diversified.
2.00
net PF (CI-LB 0.99)
n
58
58
WR
55.2%
55.2%
PF (2× cost)
1.88
1.88
Top symbol
XOM (21%)
XOM (21%)
Days
51
51
CI-LB
0.99
0.99
🔧 Why rs-breakout-scout could work for EQUITY
Relative-strength breakout: buy stocks that are outperforming their sector over 2-3 weeks, entering
on the next day's open. Works because institutional money flows create momentum that retail can ride.
Problem: n=58, CI-LB 0.99 — borderline. Needs n≥100 forward.
🏛️ 13F institutional flow — mutual-fund angle for EQUITY
A mutual-fund company doesn't need a per-pick alpha signal. It can follow institutional flow:
| Symbol | Action | Filings | Avg Value | Fundamental story |
|---|---|---|---|---|
| INTC | NEW (4 funds) | 4 | $1.85B | Intel turnaround: 4 hedge funds initiated in Feb-Apr 2026. AI-chip rebound narrative (Gaudi 3 competitive vs NVDA). |
| XOM | NEW (2 funds) | 2 | $16.0B | Energy rotation: oil-price stability ($70-80/bbl). Dividend yield ~3.2%. Hedge against inflation. |
| MSFT | HOLDING | 2 | $761B | Cloud (Azure) growth + dividend aristocrat. Consistent institutional confidence. |
| GOOG/GOOGL | HOLDING | 2 | $740B/$513B | Search advertising stability + Cloud Platform growth. Alphabet restructuring paying off. |
| META | HOLDING | 2 | $661B | Meta's ad-tech + VR/metaverse investment. 2026 recovery from 2022-23 lows. |
| NVDA | HOLDING | 2 | $594B | AI chip dominance. But concentration risk: 80% of AI-chip revenue. High valuation. |
Mutual-fund approach: Build a diversified portfolio of stocks with 2+ funds adding positions
in the latest 13F cycle. Screen for: (1) new institutional buying, (2) positive analyst revisions,
(3) positive earnings surprises. This is factor-based long-term investing, not day-trading.
3. Other classes — failed or refuted
COMMODITY
FAIL
HG=F copper: 751/815 signals overlap. Hard-only n=137 PF 1.21 → neg. after costs. 6/6 agents REFUTED.
FOREX
FAIL
forex_trend_aligned n=50 (need 100). 99.8% copy-trader contamination. EURGBP=X frozen since Jun 21.
BOND
FAIL
n=27, WR 22.2%, PF 0.85. "PF 1.79" was from replay-artifact file (MISPRICED_ENTRY rows).
FUTURES
FAIL
n=90, PF 0.37. No edge on hard-only. Thin data.
MEMECOIN
FAIL
extreme_fear n=15121 but 1 distinct day = frozen-exit artifact. 0% P&L rows: 3 (sentinels).
ETF (directional)
FAIL
n=27, WR 14.8%, PF 0.59. Toxic — no edge whatsoever.
4. Shippable products (not per-pick winners)
🟡 Tactical Rotation TAA
RESEARCH-VAL
Sharpe 1.48 vs SPY 1.37. Monthly rotation, 3-5 ETFs.
ℹ️ What this is (ELI5)
Every month, rank asset classes (stocks, bonds, commodities, REITs) by 12-month performance.
Buy the top 2-3. When one crashes, rotate to the next. This is "don't put all eggs in one basket
and move them when the basket starts shaking." Not a stock-pick — a portfolio-level product.
🔵 Mutual Fund Ranker
RESEARCH-VAL
13F flow + analyst ratings + survivorship-free.
ℹ️ What this is (ELI5)
Track where hedge funds are putting money (13F filings), then rank funds by: (1) new institutional
buying, (2) analyst upgrade ratio, (3) 12-month performance vs category. A mutual fund company
can use this to decide which funds to recommend or hold in their own portfolio.
5. Data quality issues found (stale, 0% P&L, sentinels)
⚠️ Issues discovered during cross-check
| Issue | Count | Impact | Status |
|---|---|---|---|
| 0% P&L rows (sentinels) | 3 | Hard outcomes with exactly 0% — possible sentinels | Excluded from analysis |
| Old picks (>2 years) | 270 | Back-tested on 2024-era data; market regime may be different | Flagged for review |
| Copy-trader feed frozen | ~283 trades | FOREX class PF inflated 99.8% by frozen EURGBP=X (since Jun 21) | FAIL — excluded |
| 15-minute stale pages | 32+ | /audit links to pages last updated Mar-Jun 2026 (stale) | STALE — needs refresh |
| Database connection errors | 5 endpoints | /findmutualfunds/api/* returns "Database connection failed" — PHP credentials not updated | BROKEN |
6. Decision: Should we make a pick, fine-tune, or meta-research?
🎯 Recommendation: Meta-research (fix the ruler) → Then crypto_short_rsi3050 forward test
| Path | Action | Value | Status |
|---|---|---|---|
| Meta-research | Re-resolve all history from 1h/5m bars; merge artifact detector + walk-forward to main; fix stale PHP API creds | HIGHEST | 2,000+ strategies need re-measuring |
| Crypto forward pilot | Seal crypto_short_rsi3050 (SHORT only, no leverage, 2% position cap) for 100 forward hard outcomes | HIGH | Strongest signal; needs MaxDD control |
| Ship as smart-beta ETF product (not per-pick trading) | MEDIUM | Ready to productize | |
| Build INTC/XOM 13F tracker as mutual fund allocation tool | MEDIUM | Real institutional data | |
| Trade crypto_short_rsi3050 SHORT | NOT READY | MaxDD 49% = catastrophic for institutional capital |
For a mutual-fund/ETF company: Don't trade crypto_short_rsi3050 directly.
It has a real statistical edge (PF 2.16, Sharpe 6.9) but the 49% drawdown is too volatile.
Instead: (1) ship the Tactical Rotation TAA as a product (it beats SPY risk-adjusted),
(2) build the 13F mutual-fund ranker (follows where institutions put money), and
(3) keep crypto_short_rsi3050 as a tactical overlay — a 5% portfolio allocation
that activates only when BTC funding < 5% and spot volume > 20th percentile.
This limits drawdown while capturing the edge.
7. Full per-candidate scorecard (Grok's 13-gate framework)
All candidates ranked (highest score first)
| Candidate | n | PF net | CI-LB | Sharpe | MaxDD | Score | Verdict | ~120 mo | 1.84 | ~1.4 | 1.48 | ~20% | ~55 | RESEARCH-VAL |
|---|---|---|---|---|---|---|---|
| crypto_short_rsi3050 | 125 | 2.16 | 1.43 | 6.90 | 49% | ~65 | CANDIDATE |
| MBS+200d ETF | 487 | 1.84 | ~1.4 | — | ~20% | ~55 | RESEARCH-VAL |
| IBS<0.15+200d | 775 | 1.63 | — | — | — | ~50 | RESEARCH-VAL |
| rs-breakout-scout (stock) | 58 | 2.00 | 0.99 | — | — | ~30 | CANDIDATE (weak) |
| high_trust_momentum | 54 | 1.29 | <1.0 | — | — | ~25 | CANDIDATE (weak) |
| luxalgo_confluence SHORT | 104 | 1.20 | ~1.0 | — | — | ~30 | CANDIDATE (weak) |
| copper HG=F | 137 | <1.0 | — | — | — | 0 | FAIL |
| mega_mutation | 38 | 1.73 | CI crosses 0 | — | — | 0 | FAIL (R3 NULL) |
| Copy-trader (FX) | ~283 | 0.85 | — | — | — | 0 | FAIL (frozen) |
| Kimi BOND tournament | 546 | 1.79 (raw) | — | — | — | ~20 | FAIL (60% excluded) |
| HyroTrader 35 strats | ~50 ea | 1.05-2.15 | — | — | — | ~10 | COIN-FLIP |
8. Cross-agent consensus
All 5 agents agree: 0/9 classes are PROVEN
| Agent | Best candidate | Verdict |
|---|---|---|
| Kilo (this report) | crypto_short_rsi3050 (PF 2.16, Sharpe 6.9) | CANDIDATE (MaxDD 49%) |
| Buffy | luxalgo_confluence SHORT | CANDIDATE (forward CI crosses 0) |
| INCEPTION | 0 winners, 3,446 cleaned rows, walk-forward harness | Zero classes pass strict gate |
| GHCOPILOT | luxalgo SHORT, copper HG=F | REFUTED / forward pending |
| Grok | MRBB+200d ETF, IBS+200d, crypto_short_rsi3050 | CANDIDATE (MRBB ~55pts, rsi3050 ~65pts) |
| OpenCode | high_trust_momentum | REFUTED — peers say CANDIDATE |