CVP-HDDQN: Research software for hierarchical reinforcement learning in gold futures

Abstract

CVP-HDDQN is the software companion to the manuscript Auction-state representation and expiring advice for hierarchical reinforcement learning in gold futures. Version 1.0.0 is a code-only release containing the original model, feature, learner and execution-ledger modules, selected historical Python implementation and synthetic test sources, packaging metadata and usage documentation. The architecture combines a 15-minute information publisher with a five-minute inventory-aware trader using an auction-state representation and expiring advice. The installable core supports synthetic interface checks and a forward-pass example. Historical pipeline sources require separately supplied market data and original bound contracts and artifacts. This archive contains no experimental results, performance tables, recorded policy outcomes, generated figures, raw market data, trained checkpoints or manuscript. No new financial experiments, fitting, market replay or resampling were performed for this release. Author: Arman Salehi. Code is licensed under PolyForm Noncommercial 1.0.0; project documentation is licensed under CC BY-NC 4.0. Commercial use is not licensed. Third-party dependencies retain their own licenses. The source archive, Python wheel and SHA-256 checksum file accompany this record. GitHub repository; version 1.0.0.