Original research hypothesis: test whether the signs of the last two non-overlapping 24-hour returns identify persistent differences in subsequent 48-hour returns, independently on four lightly tested Hyperliquid perpetuals using 4-HOUR bars. Learn four conditional means from matured historical labels with shrinkage, subtract unconditional drift, and trade only large forecasts. This is a discrete conditional-response model, not a prescribed candle pattern, continuously rebalanced panel or a claim that momentum always continues. No preliminary return study was performed; capture, frequency and performance figures below are predictions. Market orders execute at the completed signal bar close, the capability registry equivalent of next-bar-open execution on a continuous venue. Protective stops are platform-matched intrabar stops. Use all actually available history per instrument, admitting each only after its listing and complete warm-up; do not truncate every leg to the newest listing. Official base-tier perpetual taker fee reference: [Hyperliquid fees](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees), 0.045% per fill, 0.09% round trip. Rolling estimation is part of the frozen algorithm, not discretionary recalibration or a search on live results.