E06 is the only strategy that ever passed the lab gate. This page re-examines the three things that most often make backtest results look better than reality: a universe selected with hindsight, correction for multiple trials, and stability across periods.
E06 is downgraded from "pass" to "fail". Not because of a single p-value, but because of three mutually supporting pieces of evidence: its edge disappears once the universe is selected without hindsight, all of its profit is concentrated in 2020-2021 and 2023, and the effect vanishes when the data is sampled weekly. DSR itself cannot be used as the deciding factor because its result moves from 0.09 to 0.99 merely by changing the assumption about correlation between trials.
E06 selected BTC, ETH, and SOL. All three are still alive and have at some point risen hundreds of times over. That is a decision that could only be made after knowing the outcome. To test it, the universe is determined by a rule that does not look into the future: at the start of each month, take the 10 assets with the highest 30-day median dollar volume, with at least 180 days of history, then go long only those whose 60-day momentum is positive. The candidates include assets that are already dead or permanently collapsed (LUNA, FTT, SRM, WAVES, ANT, TORN, KLAY, BTS, XEM, REN).
| Variant | CAGR | Sharpe | Max DD | Time in market |
|---|---|---|---|---|
| BTC/ETH/SOL fixed (original E06) | 15.4% | 0.54 | -76.4% | 67% |
| Top-10 point-in-time liquidity, including dead assets | -2.8% | 0.20 | -78.3% | 92% |
| Top-10 point-in-time liquidity, only assets still alive | -3.5% | 0.19 | -80.2% | 91% |
| Buy and hold BTC | 39.2% | 0.87 | -76.6% | 0% |
With a realistic universe, this strategy does not make money: CAGR -2.8% and Sharpe 0.20, while buying BTC and then doing nothing yields 39.2% and Sharpe 0.87. The difference between rows two and three is small (0.20 vs 0.19), so what kills it is not the dead assets themselves, but the expansion of the universe: once we no longer pick the three winners, its edge disappears.
DSR answers: after trying so many combinations, how confident are we that the best Sharpe is not luck. The formula follows Bailey and Lopez de Prado, and I have rechecked the implementation: SR0 uses the expected maximum of N normal trials, the denominator uses skewness and kurtosis, and Sharpe is calculated daily before being compared. What is problematic is not the formula, but its assumptions: trials are assumed to be independent, even though our 87 runs use the same data.
| Correlation between trials | Effective N | SR0 | DSR |
|---|---|---|---|
| 0.00 | 87.00 | 0.0977 | 0.094 |
| 0.30 | 3.25 | 0.0358 | 0.946 |
| 0.50 | 1.98 | 0.0200 | 0.991 |
| 0.75 | 1.33 | 0.0021 | 0.999 |
| 0.90 | 1.11 | -0.0115 | 1,000 |
This means DSR is not final evidence in either direction. Under the independent assumption the result is 0.09 (fail), under the fully correlated assumption the result is 0.99 (pass). All that can be concluded is: E06's daily Sharpe (0.070) is below SR0 under the independent assumption (0.098) but above it under the correlated assumption. Therefore the decision on this page does not depend on DSR.
Parameters are fixed at 60 days, chosen in advance, not from holdout results. Rolling 12-month window, 3-month step.
| Window | Return | Sharpe | Max DD |
|---|---|---|---|
| 2020-10-16 → 2021-10-10 | +1129.2% | +3.55 | -32.8% |
| 2021-01-14 → 2022-01-08 | +525.3% | +2.68 | -32.8% |
| 2021-04-14 → 2022-04-08 | +107.7% | +1.42 | -33.8% |
| 2021-07-13 → 2022-07-07 | +66.3% | +1.27 | -42.2% |
| 2021-10-11 → 2022-10-05 | -35.6% | -0.91 | -54.3% |
| 2022-01-09 → 2023-01-03 | -38.9% | -1.61 | -50.0% |
| 2022-04-09 → 2023-04-03 | -14.8% | -0.18 | -43.2% |
| 2022-07-08 → 2023-07-02 | +0.7% | +0.23 | -34.2% |
| 2022-10-06 → 2023-09-30 | +11.8% | +0.48 | -27.7% |
| 2023-01-04 → 2023-12-29 | +154.5% | +2.19 | -29.7% |
| 2023-04-04 → 2024-03-28 | +177.2% | +2.43 | -29.7% |
| 2023-07-03 → 2024-06-26 | +107.3% | +1.74 | -31.3% |
| 2023-10-01 → 2024-09-24 | +93.4% | +1.63 | -44.2% |
| 2023-12-30 → 2024-12-23 | +29.6% | +0.79 | -45.6% |
| 2024-03-29 → 2025-03-23 | -28.6% | -0.71 | -45.6% |
| 2024-06-27 → 2025-06-21 | -0.7% | +0.16 | -23.0% |
| 2024-09-25 → 2025-09-19 | +73.1% | +1.52 | -23.0% |
| 2024-12-24 → 2025-12-18 | -1.0% | +0.16 | -22.8% |
| 2025-03-24 → 2026-03-18 | -0.5% | +0.16 | -31.8% |
| 2025-06-22 → 2026-06-16 | -4.5% | +0.03 | -35.3% |
| 2025-09-20 → 2026-09-14 | -12.5% | -0.31 | -35.7% |
| 2025-12-19 → 2026-10-25 | +7.8% | +0.44 | -17.9% |
The pattern is clear: Sharpe 3.55 in 2020-2021, negative throughout 2022, back to 1.6-2.4 in 2023, then approaching zero since late 2024. There is no evidence that the edge persists across different market regimes; what exists is two bull periods that cover up the rest.
| Asset | Sharpe | Return | Max DD |
|---|---|---|---|
| BTC | 1.00 | +114% | -25.8% |
| ETH | 1.01 | +188% | -20.4% |
| SOL | 1.21 | +508% | -38.8% |
Sharpe is similar across the three assets (1.0 to 1.2), so it is not one magic coin. But the return is dominated by SOL (+508%), and that is a reminder that the final result depends heavily on one asset that happened to be chosen.
| Strategy | Start | Return | Position | Days |
|---|---|---|---|---|
| E06:basket_mom | 2026-10-25 | +0.00% | cash | 1 |
| E03:donchian | 2026-10-25 | +0.00% | cash | 1 |
| E10:ma_cross | 2026-10-25 | +0.00% | cash | 1 |
E06 is not continued as a candidate. Paper trading continues as an observation record, not as a strategy expected to be profitable. The next approach most worth testing is carry funding (E17): a strategy that does not depend on price direction and has different economics, with a walk-forward Sharpe of 2.30 that should actually be treated with suspicion because it is too good. The first test for E17 is to ensure its data is not leaked and its funding costs are real, not to add a new strategy.
Updated 25 Oct 2026 08:12 UTC+7. Data sources: Binance klines and CoinGecko (local cache), recalculated on every audit. No real-profit claims are made on this page.