Hyperliquid ALGO/ICP Causal Cluster-Conditional 72-Hour Forecasts

Hypothesis

Derived by the Research Lead from Dmitriy Gizlyk’s [Practical use of clustering](https://www.mql5.com/en/articles/10943): form price/activity clusters without return labels, then estimate subsequent returns on a separate chronological sample. The source supplies this modeling approach, not these trading rules or evidence of profitability. Trade ALGO and ICP independently on 4-HOUR bars, long, short or flat. Market orders at signal-bar closes implement the registry’s faithful equivalent of next-bar-open execution on continuous markets. Use all available history separately for each instrument after causal warm-up; do not assert equal listing dates or a full 2.3-year span. The fixed algorithm refreshes learned states every 30 UTC days; frozen operation means its rules and hyperparameters remain fixed, while this prescribed causal state update continues. No pre-study or performance-based screening was run.