Changelog

New Products

• Feature Saturna — Model calibration maintenance layer for MLOps. Maintains model accuracy without retraining.
• Feature Command Dosage Control — Robotic motion smoothing through gradual command injection. Patent pending.

Optimization Engines

• Improvement QuantumJolt: 15% faster convergence on 500D+ problems through improved SPSA scheduling
• Improvement Spiral Swarm DSS: Added β-adaptive contraction for automatic explore/exploit balancing
• Feature Kestrel Pro: Added equality constraint support via augmented Lagrangian method

API Changes

• Breaking optimize() now returns OptimizationResult object instead of tuple
• Improvement Added callback parameter for progress tracking
• Improvement Added options dict for method-specific parameters

Bug Fixes

• Fix Fixed numerical instability in DSS with very small search radii
• Fix Fixed constraint violation reporting in Kestrel multi-objective mode

Features

• Feature Kestrel Pro: Multi-objective optimization with NSGA-II inspired Pareto sorting
• Feature Added benchmark() utility for standardized optimizer comparison
• Feature Added standard test functions: Rastrigin, Ackley, Griewank, Rosenbrock, Schwefel, Levy

Improvements

• Improvement QuantumJolt: Automatic learning rate scheduling based on function landscape
• Improvement DSS: Fibonacci lattice now uses golden angle for better space coverage
• Improvement Documentation overhaul with new tutorials and API reference

Features

• Feature Kestrel Pro: Constrained optimization with penalty methods
• Feature Added seed parameter for reproducible runs (QuantumJolt, Kestrel)

Bug Fixes

• Fix Fixed memory leak in long-running optimizations with callbacks
• Fix Fixed bounds checking for integer-valued parameters

Features

• Feature Spiral Swarm DSS: 100% deterministic optimization with zero hyperparameters
• Feature Validation on 1000D problems showing significant advantage over SciPy

Improvements

• Improvement QuantumJolt: Improved handling of noisy objective functions
• Improvement Better error messages for invalid bounds and constraints

Initial Release

• Feature QuantumJolt optimizer for high-dimensional multimodal optimization
• Feature Unified optimize() API
• Feature Python 3.8+ support
• Feature Initial documentation and examples