Hyperliquid Log Return topic goes live on Forge for predictions

Forge: Forge serves as the model deployment and competition hub within the Allora Network, allowing ML practitioners to build, test, and promote inference workers on specialized prediction topics. It supports testnet environments like topic 90 for evaluating models before potential mainnet graduation via performance criteria. The Hyperliquid Log Return topic launched on Forge testnet to crowdsource dynamic, multi-market log-return forecasts without fixed asset lists. Hyperliquid: Hyperliquid operates as a decentralized perpetual futures exchange focused on onchain derivatives trading with its own matching engine and ecosystem expansions including HyperEVM. The platform enables permissionless market creation and supports high-leverage trading across a wide range of assets. The Hyperliquid Log Return topic on Forge directly targets short-horizon predictions for its perpetual markets, leveraging the platform's one-minute candle data for ground truth. Network Goals: The setup tests whether aggregating weakly correlated model edges across many markets can outperform individual models, with promotion to mainnet hinging on statistical confidence bounds rather than headline metrics. Evaluation Design: Scoring employs the asymmetric CZAR loss function on log returns to better reward directional skill over conservative zero predictions in noisy short-horizon environments. Prediction Innovation: The topic uses a dynamic registry rebuilt each round from model submissions, allowing coverage to shift toward markets where contributors see predictive edge rather than relying on a static asset list.