DynaSchedBench

DynaSchedBench accompanies the ICML 2026 paper DynaSchedBench: Calibrated Dynamic Scheduling Benchmarks and Observability Paradox in LLM-based Scheduling Agents. It provides calibrated Dynamic Flexible Job Shop Scheduling instance generation, an event-driven simulator, a single-agent scheduling environment, trajectory evaluation, visualization tools, and dsbx-* command-line workflows for testing heuristic, evolutionary, and LLM-based scheduling agents.

Official website: https://dsbx7.github.io/

Get Started

• Installation
  • Supported Python Versions
  • Install From PyPI
  • Install From Source
  • Optional LLM Agent Configuration
  • Verify Local Documentation Builds
• Quickstart
  • 1. Install DynaSchedBench
  • 2. Prepare an input model
  • 3. Generate an instance
  • 4. Validate the event stream
  • 5. Run a baseline agent
  • 6. Evaluate and visualize
  • 7. Use the Python API
• Concepts
  • Input Model
  • Generated Instance
  • Event Stream
  • Simulator Snapshot
  • Environment And Agent
  • Trajectory
  • Cold-Start And Warm-Start Modes
• Input Model And Instance Format
  • Minimal Input Model
  • Top-Level Sections
  • Generated Files
  • Validation
• Workflows
  • Single-Instance Experiment
  • Compare Several Agents
  • Batch Generation
  • MOO And Hybrid Calibration
  • Warm-Start Evaluation
  • Simulator Snapshot Inspection

Command-Line Workflows

• Command-Line Interface
  • dsbx-gen
  • dsbx-agent
  • dsbx-eval
  • dsbx-vis
  • dsbx-sim
• Agents
  • Discover Available Agents
  • PDR Agents
  • Evolutionary And Local Search Agents
  • LLM Agents
  • Python Agent Interface
• Evaluation
  • Trajectory Files
  • Recompute Metrics
  • Metric Families
  • Validate Generated Events
  • Check Schedule Feasibility
  • Warm-Window Evaluation
  • Debug LLM Episodes
• Visualization
  • Machine Gantt
  • Job-Centric Gantt
  • Metrics Summary
  • Metric Curve
  • Event Timeline
  • Interactive Gantt Inspector
  • Practical Tips
• Benchmarks
  • Generated Instance Directory
  • Agent Run Directory
  • Repository Data Layout
  • Reproducibility Checklist
  • Scaling Experiments
• FAQ
  • Is The PyPI Name DynaSchedBench Or dsbx?
  • Why Does dsbx-agent run Use -d Instead Of -c?
  • Should I Use trajectory.json Or trajectory_light.jsonl?
  • Why Is My Read the Docs Build Failing On A Warning?
  • Do I Commit docs/_build/html?
  • How Do I Add A New CLI Tutorial?

API Reference

• API Reference
  • Agents
  • Environment
  • Evaluation
  • Generation
  • Simulation
  • Visualization