VERIFICATION LEDGER
What we back-tested, and what actually survived
Most sites publish only the strategies they want you to trade. This is the full ledger: every systematic study we ran, scored on the same axis, failures included. Publishing the failures is the point — a method that survives 43 pre-registered tests is worth more than one that was never tested.
43
studies
4
live
28
failed outright
| What was tested | Verdict | Survivors | Out-of-sample | Deflated Sharpe |
|---|---|---|---|---|
| 24-hour drift after economic-event surprisesRead the full study | Live | 2 | — | — |
| Correlation systems: as an entry signal versus as exposure control | Live | — | — | — |
| US market close leading next-day Nikkei 225 and FXRead the full study | Live | 2 | — | — |
| Correlation atlas across commodities, indices and volatility | Live | 2 | — | — |
| Seven well-known retail methods (TD9, NR7/ORB, Turtle Soup, Gotobi, month-end rebalancing, weekend gap) | Shadow | 3 | — | — |
| An agent that watches every chart and decides trade by trade: does judgement beat the machine and a coin flip? | Shadow | 0 | — | — |
| CFTC Commitments of Traders speculative positioning | Close, not enough | 0 | 0.167 | 0.043 |
| Grid trading: evenly spaced limit orders, small take-profits, no stop lossRead the full study | Close, not enough | 2 | — | — |
| Japan MOF weekly cross-border portfolio flows and the yen | Close, not enough | 0 | 0.286 | 0.327 |
| Japan MOF monthly outbound investment by investor sector (insurers, pensions, banks, trusts) and the yen | Close, not enough | 0 | 0.333 | 0.069 |
| Extended Japan flow model: seven pairs with direction fixed by mechanism | Close, not enough | 0 | 0.833 | 0.413 |
| Interest-rate carry: BIS policy-rate ranking, eight-currency cross-sectional portfolioRead the full study | Close, not enough | 0 | 0.833 | 0.935 |
| Every technical-analysis family combined pairwise and beyond, across all 21 pairs | Close, not enough | 0 | — | — |
| Applying the KAI persona rules to all 21 pairs: which ones actually make money | Close, not enough | 2 | — | — |
| Re-ranking every live strategy by calendar-year total profit instead of per-trade edge | Close, not enough | 0 | — | — |
| Standard technical indicators (moving averages, MACD, RSI, Bollinger, ADX, currency strength) | Failed outright | 0 | — | — |
| Modern candlestick and chart patterns (engulfing, pin bar, Heikin-Ashi, Supertrend, Ichimoku) | Failed outright | 0 | — | — |
| Five trend-reversal detectors (MA200, MA50 slope, Supertrend, Donchian, CUSUM) | Failed outright | 0 | — | — |
| Trade-quality suite (SQN, e-ratio, holding-time optimisation) | Failed outright | 0 | — | — |
| Machine-learning meta-labelling to filter event-drift trades | Failed outright | 0 | — | — |
| Scalping across 24 micro-session cells, priced with measured spreadsRead the full study | Failed outright | 0 | — | — |
| Fibonacci combined with Elliott Wave, MACD and RSIRead the full study | Failed outright | 0 | — | 0.145 |
| Portfolio-style technical factors (time-series momentum, relative strength, low volatility) | Failed outright | 0 | — | 0.270 |
| Six widely shared social-media logics (fair value gap, order block, liquidity sweep, session-limited entries, market structure shift, first-move fade) plus lot doublingRead the full study | Failed outright | 0 | — | 0.024 |
| Commodity-currency basket against gold, silver and oil | Failed outright | 0 | 0.396 | — |
| US dollar net liquidity (Fed assets minus TGA minus reverse repo) against a dollar basket | Failed outright | 0 | — | 0.327 |
| Changes in interest rates: monetary-policy divergence, long the hiking side | Failed outright | 0 | 0.333 | 0.119 |
| A risk-off index used as a defensive gate on carry | Failed outright | 0 | 0.167 | 0.766 |
| US market close leading next-day commodities, crypto, and Asian equity indices | Failed outright | 0 | — | — |
| US-to-Nikkei transmission beyond correlation: volatility, gap-fill, asymmetry, multi-day drift | Failed outright | 0 | — | — |
| Deep-dive on the sole survivor: harmonic-PRZ times Bollinger fade under a true 1994-2009 holdout | Failed outright | 0 | — | — |
| The full candlestick canon — 50 families including Sakata methods — across three eras since 1971 | Failed outright | 0 | — | — |
| Machine learning at full power: gradient boosting and a Transformer on GPUs, with purged walk-forward | Failed outright | 0 | — | — |
| Asymmetric payoff: trailing stops and reward-to-risk targets against time exits, resolved on 5-minute paths | Failed outright | 0 | — | — |
| Separating entry from exit: edge ratio (MFE/MAE) scores entries without any exit rule | Failed outright | 0 | — | — |
| Re-screening macro and positioning signals, then testing them episode by episode | Failed outright | 0 | — | — |
| Adaptive exits that react to the trade thesis dying, versus a fixed time stop | Failed outright | 0 | — | — |
| Signal threshold versus trade frequency, judged by annual profit rather than per-trade edge | Failed outright | 0 | — | — |
| Carry revisited pair by pair: annual totals across all 21 pairs against realistic swap markups | Failed outright | 0 | — | — |
| Carry tested on CME currency futures, where there is no broker swap markup to blame | Failed outright | 0 | — | — |
| Can the strategy’s bad years be seen coming? Volatility, correlation and drawdown gates tested | Failed outright | 0 | — | — |
| The 50 candlestick families tested on gold, silver, bitcoin and stock indices: none survived, same as FX | Failed outright | 0 | — | — |
| Twelve aggressive strategy families | Failed outright | 0 | — | — |
0.50 means no better than chance. Only values above 0.50 carry information.
How these were judged
Gates are registered before the test runs, so a result cannot be reinterpreted after the fact. Out-of-sample is scored only after the direction has been fixed on in-sample data. Deflated Sharpe discounts the number of variants tried, which is what makes most published back-tests unreliable. Reversing a sign after seeing the result counts as a failure, not a discovery.
Research records, not investment advice. Past back-test or live results do not predict future returns. Trading foreign exchange carries risk of loss.
Ledger updated: 2026-08-12