AutoQuant: An Auditable Expert-System Framework for Execution-Constrained Auto-Tuning in Cryptocurrency Perpetual Futures

Battle Count: 52

Directly addresses execution-constrained configuration selection for cryptocurrency perpetual futures trading. Provides a complete framework for realistic backtesting, parameter tuning, cost modeling, funding alignment, and governance in high-friction crypto derivatives markets. Highly relevant for practitioners building algorithmic trading and risk-control pipelines in 24/7 derivatives markets.

Requires implementing strict t+1 execution semantics, no-look-ahead funding alignment, multi-scenario cost grids, Bayesian optimization (TPE via Optuna), two-stage screening with ex-ante thresholds, deterministic artifact export, full-chain accounting invariants, and guard overlays. CPU-only feasible with bounded budgets (40-120 trials). Moderate engineering burden for data plumbing, timestamp alignment, and cost accounting. The framework is signal-modular but requires careful configuration of execution rules and cost profiles.

The paper provides deterministic machine-readable artifacts (configuration JSON, per-bar ledger CSV, robustness summaries, invariant-check files), minimal scripts and commands to regenerate tables/figures, a replication package with anonymized data and scripts, third-party replay checks (Backtrader, Freqtrade), strict t+1 parity snapshots, and full-chain accounting invariants. All date ranges in UTC, bar-open timestamps, and ex-ante thresholds are explicitly documented. Raw exchange data not redistributed due to licensing but regeneration scripts provided.