Trying out k means clustering now whereby the data gets split into groups using similarity. In this case: it takes every extended asset and measures five parameters: how extended the asset is, how long it’s been there, how fast it’s moving, how rare that level is, and how much volume is behind it. Four groups emerged: Noise spike: got there fast, already moving back. Brief touch, probably not worth trading. Slow grind: been extended for multiple time cycles, low velocity. Potentially trapped positioning building. Crowded position: extreme percentile rank, moderate volume. Squeeze or liquidation risk depending on direction. Thin market — low volume relative to extension. The z-score is technically valid but needs more digging. Detailed article to follow on the entire process.