One rule set, read once a day at the 00:00 UTC close and acted on at the next open. This page reports the configuration the system actually runs: a 33.33% floor in the dominant major with satellites capped at 25% each. It also reports, at length, why the headline figure flatters itself.
Crypto trend positive, pick the stronger major, then split with whichever mid caps are beating it. Crypto negative, gold while gold is trending up, otherwise the S&P while it is trending up, otherwise cash.
The live system now asks a fifth question when everything else says no, and holds the cash in euros or dollars depending on which is stronger. The test does not. It sits in undifferentiated cash on those days.
The effect is small. Over the recent sample the ladder reached the cash rung rarely, and the euro side of it rarer still. But it means the tested system and the live system are not identical, and you should know that before reading anything below.
Month end equity on a log scale, against buying Bitcoin on day one and never touching it. Both start at $10,000.
Worst drop here is the deepest fall inside that calendar year, not the fall from the all time peak.
| Year | Start | End | Return | Worst drop | Fills |
|---|---|---|---|---|---|
| 2018 | $10,000 | $13,566 | +35.7% | −52.6% | 77 |
| 2019 | $13,566 | $27,071 | +99.6% | −40.6% | 84 |
| 2020 | $27,071 | $127,589 | +371.3% | −35.9% | 109 |
| 2021 | $127,589 | $1,034,689 | +711.0% | −44.4% | 296 |
| 2022 | $1,034,689 | $680,506 | −34.2% | −45.1% | 68 |
| 2023 | $680,506 | $2,173,172 | +219.3% | −22.4% | 203 |
| 2024 | $2,173,172 | $6,184,207 | +184.6% | −31.9% | 295 |
| 2025 | $6,184,207 | $11,548,056 | +86.7% | −21.6% | 131 |
| 2026 (to 15 Sep) | $11,548,056 | $11,024,152 | −4.5% | −21.7% | 123 |
Lowering the floor raised the final number at every single step tested, all the way down to zero, and the drawdown moved very little. A rule with no interior maximum is not being optimised, it is only being pushed further out on risk. So the choice was made on Sharpe and on what survives a weaker signal, not on the biggest number.
| Major floor | Final equity | Annualised | Worst drawdown | Sharpe | Without 2020 and 2021 |
|---|---|---|---|---|---|
| 0% (no floor) | $18,485,642 | +137.3% | −69.0% | 1.53 | 42.0x |
| 25% | $13,309,439 | +128.5% | −66.3% | 1.57 | 33.1x |
| 33.33% | $11,024,152 | +123.6% | −65.0% | 1.57 | 28.8x |
| 40% | $9,366,413 | +119.4% | −63.8% | 1.56 | 25.5x |
| 50% | $7,260,188 | +113.1% | −62.2% | 1.55 | 21.0x |
| 60% | $5,732,044 | +107.4% | −60.4% | 1.53 | 18.4x |
| 100% (no mid caps) | $1,412,028 | +76.6% | −61.3% | 1.30 | 7.3x |
Sharpe is computed on daily returns, annualised, with no risk free rate. The last column strips 2020 and 2021 out of the compounding entirely and shows what the other seven years did on their own. 33.33% is the row the system runs.
The result above only holds while the ratio TPIs keep picking winners. To find out how much of that edge can disappear before a lower floor becomes the wrong choice, the mid cap selection was replaced on a share of days by a random draw with the same hit rate. At 100% the real signal is used every day. At 0% every pick is random.
| Signal quality kept | Floor 33.33% | Floor 40% | Floor 50% | Floor 60% |
|---|---|---|---|---|
| 100% (the real signal) | $11.02M | $9.37M | $7.26M | $5.73M |
| 75% | $6.05M ($3.83M) | $5.55M ($3.69M) | $4.87M ($3.23M) | $4.13M ($2.84M) |
| 50% | $5.05M ($2.15M) | $5.04M ($2.14M) | $4.44M ($2.07M) | $3.62M ($1.98M) |
| 25% | $3.75M ($1.20M) | $3.80M ($1.28M) | $3.77M ($1.41M) | $3.62M ($1.43M) |
| 0% (fully random) | $1.93M ($852k) | $2.01M ($948k) | $1.99M ($1.06M) | n/a |
Median final equity across 40 runs per cell, 60 for the fully random row. The bracketed figure is the 10th percentile, the bad case. The real signal beat the best of 60 fully random runs at every floor, so the selection does carry edge inside this sample.
Share of the 3,180 days spent in each leg. Gold and the S&P substitute together account for 39%, and that is where the drawdown protection comes from.
| Bought on 1 Jan 2018, never touched | From $10,000 | Return | Worst drawdown |
|---|---|---|---|
| Bitcoin | $55,152 | +451.5% | −81.2% |
| Ethereum | $32,718 | +227.2% | −94.0% |
| Gold | $32,911 | +229.1% | −26.6% |
| S&P 500 | $28,209 | +182.1% | −33.9% |
A backtest that only shows its best face is worth nothing. These are the four things that would make a live result smaller than the tested one. Read them before the calculator, because the calculator will happily show you a number that carries every one of these problems inside it.
Every indicator in every one of these systems was tuned on the same history it is measured against. A tuned rule set always looks better on its own training data than it will in front of you. Treat the shape of the result as real and the size of it as optimistic.
SOL, BNB, SUI, XRP, DOGE and HYPE are six names that survived and did well. Nobody writes that list in January 2018. This survivorship is worth a meaningful share of the 2020 and 2021 numbers, and it sits in exactly the layer the floor controls.
2020 and 2021 turn $27k into $1.03m. Take those two years out and the other seven still compound to 28.8x. Strong, but not 1,102x.
0.10% per side is charged on every trade, which comes to $615,095 across 1,386 fills. Spread, perpetual funding and the cost of moving real size in SUI or HYPE are not modelled. Assume a few percent a year on top.
Every signal here is regenerated from the saved TradingView templates and compared day by day with what the live system recorded. ETH/BTC and gold matched on all 623 days. The one gap is a candle that had not closed yet.
Sixteen days across thirteen ratio systems, compared to the decimal against the live signal log.
Over the longest untouched stretch, 41 days at 100% ETH in 2020, the equity curve tracks the ETH price exactly.
Pick a date and an amount. This replays the same daily equity track from that day forward, at the 33.33% floor. It is a replay of tested history. Nobody received these amounts, and nothing here says the next eight years look like the last eight.
| Year | Value at year end | Return | Worst drop in year | Bitcoin instead |
|---|---|---|---|---|
| 2022 | $6,577 | −34.2% | −45.1% | $3,579 |
| 2023 | $21,003 | +219.3% | −22.4% | $9,149 |
| 2024 | $59,769 | +184.6% | −31.9% | $20,248 |
| 2025 | $111,609 | +86.7% | −21.6% | $18,965 |
| 2026 to 15 Sep | $106,546 | −4.5% | −21.7% | $16,368 |
$10,000 invested at the open on 1 Jan 2022 becomes $106,546, a 10.7x multiple over 4.7 years. The same money in Bitcoin, bought and held, becomes $16,368. Costs of 0.10% per side are already taken out of the ladder figure. This is a replay of tested history, not a projection, and not an amount anyone received.