This tests a mechanism the factory has already promoted, but on a different exchange it has barely used. The promoted Binance strategy (SpotPerpVolumeShareCashAccumulationAltBasketLong: 213 trades, average +9.78% per trade, walk-forward OOS Sharpe 1.70) buys an alt on spot when spot's share of spot+perp volume is unusually high and the coin is in an uptrend. Its claim is that a rise paid for with cash lasts, while one paid for with leverage gets unwound. If that is really about who is buying, and not a quirk of Binance's venue mix, it should also show up in OKX's own order flow, which has a different user base and a different spot/swap ratio. Here the signal and the trades both use OKX data only: share = OKX spot base volume / (OKX spot + OKX USDT-swap base volume), traded long-only on OKX_SPOT with a 1x cash account. Fixed universe, chosen by a history rule before any returns were looked at: every non-BTC/ETH alt with both OKX spot and swap daily history starting on or before 2021-01-31. That gives SOL, XRP, ADA, DOGE, LINK, LTC, DOT, AVAX, ATOM, BCH, ETC and FIL; weak names stay in. PRE-STUDY (OKX public REST daily candles, 2020-01 to 2026-10, gross of costs, close-to-close): unconditional 5-day forward return +0.84% mean / -0.33% median. Trend-only control +1.70% / -0.27%. Joint rule (share z>1.6 and 77-day return >0) +3.62% / +1.21% over 1,205 coin-days, so the share filter roughly doubles the trend-only mean. Taking one position per coin at a time with a 5-day hold gives 525 events, mean +2.56%, median +0.55%, 52% win rate. By year: 2020 +7.5%, 2021 +4.8%, 2022 -1.5%, 2023 +2.9%, 2024 +3.6%, 2025 -1.7%, 2026 +2.0%. 10 of 12 coins were positive (ATOM and ETC slightly negative). Weakness, stated up front: it loses in alt bear years, and the holdout covers mostly 2025. Fees: about 0.25% round trip (0.20% taker plus impact) against about 2.5% gross per event. Execution: market orders at the daily signal close. On a 24/7 venue that price equals the next bar's open; this is the registry's use_equivalent for next_bar_open_fill. The developer must use O(1) rolling updates, because 24 daily series is about 55k bars. Expose require_share_filter (default true) so the matched trend-only control can be run with one flag.