Nothing here tells you what to buy or sell. This piece documents a rule change to our candidate strategy, the evidence behind it, and the caveats that travel with it. The change is being tested forward in public, with no capital behind it. Nothing here is investment advice. Decisions are yours.
Yesterday we adopted the third dated amendment to our candidate strategy. It is one clause long: when Bitcoin's implied volatility is in the top of its historical range, the short book stands down. We call it the capitulation floor. It is now running on the shadow page, marked daily with no capital, next to everything else we publish. This is the story of where the clause came from and what it had to survive before we let it in.
One clause, not a rebuild
The short side of the candidate already had two conditions: a downtrend that has held for five days, and a floor on the Fear & Greed index, because shorting into extreme fear tends to mean shorting the bottom. Version 2.3 adds a third: the strategy also refuses to short when the Deribit DVOL index, the market's own estimate of coming Bitcoin volatility, sits above roughly the 36th percentile of everything it has done before that day. On the quiet, grinding stretches of a bear market that number is low and the gate behaves exactly as before. In a capitulation, when options traders are pricing panic, it spikes, and the strategy waits. Everything else, the selection, the sizing, the exits, the three staggered entry cohorts, is unchanged. If the volatility feed ever goes dark the clause simply switches itself off and the gate falls back to the previous rules, so the failure mode of the new dependency is the old strategy, not a broken one.
It came out of a study that mostly failed
We did not go looking for a volatility rule. In late July we finished collecting five years of derivatives data for the whole perpetuals universe, open interest above all, and ran a pre-registered screen over it. The headline result was a lesson in humility: open interest carried real, stable ranking information across five straight years, and none of it survived contact with the actual strategy. Every attempt to use it to veto short candidates lost to a control that dropped candidates at random. Real signal, no usable edge. We wrote it down and moved on.
The side finding was the interesting one. As part of the same screen we tested implied volatility head to head against the Fear & Greed index at predicting broad altcoin returns, and DVOL won at every horizon we measured, with a sign that surprised us: high implied volatility preceded bounces, not further losses. That is the signature of capitulation. Panic-priced options mark the moment sellers are exhausted, which is close to the worst moment imaginable to be opening new shorts. Our own trade history agrees: the forced liquidations in the short leg's honest accounting cluster in exactly these windows, and squeeze risk is the one failure mode a five-day confirmation cannot see coming.
A threshold we refused to tune
The obvious way to ruin a finding like this is to sweep the threshold until the backtest looks best. We did not sweep it at all. The Fear & Greed floor already blocks a known fraction of gate days, just under a fifth of them over the comparable window, so we set the volatility threshold to block exactly the same number of days and locked it before running anything. The new clause was forced to be an exposure-matched sibling of a rule we already trusted: same amount of standing aside, different choice of which days. Whatever the test then showed could only come from picking the days better, not from shorting less.
The test it had to survive
The amendment had to pass five pre-registered gates against the incumbent, and the decisive one was a placebo: forty runs in which the same volatility series, cut into two-month blocks, was shuffled in time and the whole backtest rerun. A rule that merely benefits from standing aside sometimes would score no better than its shuffled ghosts. The real rule beat the best of the forty, in both placebo constructions we ran. On the short sleeve itself the result was not subtle: the worst-case drawdown roughly halved and the risk-adjusted return roughly doubled, with the improvement spread across 2021 through 2025 rather than resting on one lucky year. The combined record's backtest, out-of-sample window first as always, is on the record page, regenerated under the new rule with its full ledger.
One variant failed, and the failure taught us as much as the pass. We also tested replacing the Fear & Greed floor with the volatility floor outright, on the theory that a market-derived options index should dominate a survey-flavored sentiment score. It did not. The swap looked spectacular in aggregate and fell apart under year-by-year scrutiny, because the two floors turn out to stand aside on different days: one catches the extreme-fear bottom bounce, the other catches the high-volatility squeeze. They are complements, not substitutes. The fear floor stays.
What we do not know yet
The honest list. DVOL only exists from spring 2021, so the clause has been examined against a single full bear cycle; before late 2021 the amended strategy and the old one are the same strategy by construction. In 2026 so far the clause has been a small net cost, not a gain, and we adopted it anyway because five years of evidence outweigh seven months. And it was one of several ideas we tested that day, most of which we rejected; the trial count is disclosed in the pre-registration, because a rule that emerges from many attempts deserves more suspicion than one that arrives alone. The backtest multiple is not the forward expectation. What we expect, if the evidence holds, is a strategy that gives back less in squeezes. What we published is a record that will say so either way.
In shadow now, on a record that never restarts
The amendment went live in the shadow on July 30, 2026, as a dated segment on a record that is never restarted and never backfilled. The days already published stay exactly as they were published; positions that were open under the previous rules were closed at their last published marks and labeled with the rule change in the shadow's event log, so the history reads the way an account statement would. Since-inception performance spans the rule change on purpose. A record that only counts its newest ideas is a marketing document; ours includes every version we were willing to run in public, which is the only version of honesty we know how to automate. No capital follows any of it. Decisions are yours.