Estimated by the algorithm, the algorithm assumes that the current total position is evenly distributed across common leverage levels (e.g., 5x, 25x...100x) and long/short directions. More advanced models perform multiple fittings using historical liquidation data streams to infer the approximate leverage distribution profile of the market (e.g., estimating that 20% of people use 100x leverage, 35% use 20x leverage, etc.). By comparing actual liquidation orders (provided via exchange APIs) with the algorithm, the model can be continuously refined, making the algorithm's estimated results closer to the actual liquidation outcomes.