Eaten by frictionRejected · long / short · n/a — no backtest
The idea
When a leveraged perpetual-futures position falls below maintenance margin, the exchange closes it at market. The trader is not asked. These orders do not care about price, and they come in clusters: one liquidation moves the price, which triggers the next. The bet is that cascades overshoot and then revert, and that a patient trader on the other side gets paid for providing liquidity.
Of all the candidates in the registry, this has the cleanest obligation. There is no step at which the trader can decline.
How it was tested
It never reached a backtest. The pre-run screen asks whether the trade can pay for its friction at all. Crypto friction is 0.130% round trip; spread over the declared holding of 1-5 days, even the longest holding needs 0.078% of alpha per day. The lab's demonstrated ceiling was 0.073%, later rescaled for Bitcoin's volatility to 0.072%. Clearing it would take a holding of 5.34 days, outside the declared range.
Where it died
It failed the screen when it was queued on 2026-09-08. Data was a second blocker: the large exchange kept open-interest history for only a rolling 30-day window.
The re-check was set up to avoid the obvious trap. A volatility-scaled ceiling was derived and pre-registered on 2026-09-13, before looking again. It showed Bitcoin at about the same volatility as equities, not several times higher, so the ceiling barely moved. On 2026-09-29 a liquidation history turned up on a decentralised exchange's public indexer, and the screen was re-run: 0.078% needed against 0.072%. One blocker cleared, the other confirmed. It was rejected on 2026-09-30.
The record
Thesis
When a leveraged perpetual-futures position falls below maintenance margin, the exchange's liquidation engine closes it at market without consulting the trader. Clusters of these forced, price-insensitive orders should overshoot and then revert. Of all the candidates in the registry this has the cleanest obligation: there is no step at which the trader can decline.
Gates
Gate F — Frictionfail
Is the alpha per trade large enough to pay for its own execution?
required alpha per day ≤ demonstrated ceiling, at the declared 1-5 day holding
round trip friction pct+0.13required alpha per day pct+0.08ceiling per day pct+0.07min viable holding days+5.34
Pre-run screen, not the post-run 0.30% rule. At the longest declared holding (5 days) it needs 0.078% per day; the ceiling, rescaled for Bitcoin's volatility, is 0.072%. The shortest holding that would clear is 5.34 days, outside the declared range.
Why it died
The obligation is real; the trade cannot pay for itself. Proportional crypto friction is 0.130% round trip, and spread over a holding of at most five days that needs more alpha per day than this lab has ever demonstrated. It failed the screen when it was queued on 2026-09-08 (0.078% needed against an unscaled 0.073% ceiling) and again on 2026-09-29 against a ceiling derived from Bitcoin's measured volatility, which turned out to be about equal to equities', not three to five times higher. With the data blocker gone and the friction blocker confirmed, nothing was left pending.
Open questions
Ether alone runs at higher volatility, with a ceiling of 0.116% per day that would clear the requirement. Reaching for it right after the pooled figure failed is the flattering-guess shape the screen exists to prevent; an Ether-only candidate would need its own fresh pre-registration.
What it taught
A strong obligation does not guarantee a tradeable edge: the friction screen runs independently of the mechanism's quality. The volatility-scaled ceiling was first thought of after seeing this failure, so it was not applied then; it was derived separately and pre-registered before re-checking, and it confirmed the failure rather than rescuing it. Unblocking the data (a liquidation history on a decentralised exchange, found after the large exchange's history proved limited to 30 days) changed nothing about the verdict.
Verdict history
2026-09-08Queued as a stronger mechanism than the crypto momentum candidate then being pre-registered; friction screen fails at every declared holding Eaten by friction
2026-09-08Data blocked — open-interest history limited to a 30-day rolling window
2026-09-13Volatility-scaled crypto ceiling derived and pre-registered before re-checking — Bitcoin about 1.0× equity volatility
2026-09-29Data unblocked on a decentralised exchange's public indexer; friction screen re-run, still fails (0.078 vs 0.072% per day)
2026-09-30Rejected — both blockers resolved, one favourably and one not Eaten by friction