The earlier volatility-sizing result weakened when we accounted for missing price observations. Our final bounded experiment asked a different question: could we keep Bitcoin exposure and borrow only when the existing trend rule was positive?
The rule was fixed before the results
Target Bitcoin exposure was 1× account equity, rising to 1.5× when the existing breakout rule was long. The central channel uses 120 four-hour observations to enter and 60 to exit: nominally 20 and 10 days. Targets update at actual UTC-midnight closes. The nine neighboring channel settings were also fixed beforehand.
This is an equity-exposure target, so rebalancing can change the number of coins. Borrowing is hypothetical. We compared the rule with buying Bitcoin once, fixed 1.25× and 1.5× exposure, an inverse trend control and a week-old trend control.
The tables use 6 basis points of cost per amount traded, four-hour execution delay and 10% annual borrowing cost. Each account starts with 10,000 native quote units. USD and USDT results use their own prices without currency conversion. Historical windows end in April 2026; the recent replay is a separate frozen June–September dataset. These are historical simulations, not the public scoreboard or a live record.
Since 2019
| Strategy | USD source | USDT source |
|---|---|---|
| Bitcoin core + trend exposure | +4,807.22% | +4,678.12% |
| Buy and hold Bitcoin | +1,975.43% | +1,982.08% |
| Fixed 1.25× exposure | +2,226.97% | +2,240.89% |
| Fixed 1.50× exposure | +2,059.39% | +2,074.99% |
| Extra exposure against the trend | +772.22% | +802.56% |
| Week-old trend timing | +4,659.33% | +4,016.43% |
This period supports the hypothesis historically. The central rule beat Bitcoin and both fixed leveraged controls under all twelve cost, delay and borrowing assumptions on both sources. All nine neighbors beat those controls at the primary settings. The result is worth preserving.
Since 2023
| Strategy | USD source | USDT source |
|---|---|---|
| Bitcoin core + trend exposure | +528.84% | +538.21% |
| Buy and hold Bitcoin | +365.18% | +367.89% |
| Fixed 1.25× exposure | +458.87% | +463.43% |
| Fixed 1.50× exposure | +540.61% | +547.54% |
| Extra exposure against the trend | +362.16% | +362.99% |
| Week-old trend timing | +349.44% | +353.37% |
The newer historical period is weaker. Fixed 1.5× exposure beats the central rule. The center beats all three return controls in only five of twelve assumptions, and only two of nine neighbors do so at the primary settings.
June–September 2026
| Strategy | USD source | USDT source |
|---|---|---|
| Bitcoin core + trend exposure | +32.28% | +31.30% |
| Buy and hold Bitcoin | +26.19% | +26.06% |
| Fixed 1.25× exposure | +32.26% | +32.09% |
| Fixed 1.50× exposure | +38.34% | +38.14% |
| Extra exposure against the trend | +31.74% | +32.40% |
| Week-old trend timing | +23.60% | +24.25% |
Recently, fixed 1.5× wins again. The inverse trend control also beats the central rule on the USDT source. Borrowing more can raise returns when Bitcoin rises; this alone does not establish useful timing.
The registered screen failed
Before seeing results, we required at least two of three periods—since 2019, since 2023 and the recent replay—to qualify on both sources. A period had to beat Bitcoin and fixed leverage at primary settings, win under at least eight of twelve cost/delay/borrowing assumptions, have at least six of nine successful neighbors, and pass the inverse/week-old timing comparisons. Every central account also had to remain solvent.
Only the since-2019 period qualified: one of three, with two required. Since 2023 had five of twelve all-control wins and two of nine qualifying neighbors. The recent period had zero of either. We did not change the rule, the threshold or the chosen central setting after reading the results.
The losses were severe
The central rule's primary drawdown since 2019 was about 83% on both sources. From a 2018 start, it reached 89.36% on USD and 85.49% on USDT. Across all central scenarios, the worst drawdowns were 90.14% and 85.86%. A large eventual return does not erase the experience of those losses.
In 2022, the central rule lost about 71% versus Bitcoin's roughly 65% loss. It also lost more than Bitcoin in 2018, 2025 and partial 2026. These adverse years and all neighboring results remain in the downloadable data.
Drawdown is measured at observed closes; losses inside candles or missing intervals could be worse. No forced sale or insolvency occurred in these modeled accounts, which does not validate actual margin safety or financing availability.
Where we stop
This candidate is not promoted. As agreed before the result, we are stopping new strategy searches and keeping the existing paper trial and benchmark. A separate, subsequently requested diagnostic examines whether the original long/short entries have systematic timing errors. That analysis does not change this failed verdict or enable automatic learning.
Complete evidence
There are 5,040 account calculations: 3,024 candidate cases, 672 timing controls and 1,344 repeated fixed-control verification cases. They are correlated scenarios, not independent trials. All repeated controls and 23,600 original decisions match exactly; the 588-test suite passed for this study. Software correctness is separate from trading profitability.
Both sources retain missing observations. The existing channel counts observed bars, which can cover extra calendar time across gaps. The historical periods have already been explored; no unseen market holdout is claimed.