Skip to content

Roadmap

What is planned, what was considered and deliberately left out, and why. Suggestions are welcome as issues (there is a Model request template for new built-in models).

Next

  • Faster core. The process engine takes about 1.7x as long as SimPy on a like-for-like M/M/2 queue (see the benchmarks). A profile shows no single hot spot left: the gap is spread over the statistics simulsi collects by default on every grant and release (time-weighted busy units and queue length, waits), the process bookkeeping that makes deadlock warnings and checkpoints possible, and the event queue. The package stays pure Python; a compiled extension (Cython or mypyc) would complicate wheels for every platform and the in-browser demo, so it is only an option if users hit real limits.
  • Ray backend. Experiments run on any concurrent.futures executor (local processes or Dask). Ray has no such executor built in, so it needs a small adapter and real cluster testing.
  • Flowchart view. A visual view of YAML flowcharts in the dashboard.
  • Crew pairings across days. The airline twin simulates one day, so crew rest rules that span nights are out of its reach today.

Done recently

See the changelog: event-driven wait_until(on=[...]); continuous Levels with exact threshold crossings; movement on networks and grids (simulsi.spatial); batch and parallel (split/join) stations in YAML flowcharts; experiments on Dask clusters; animated GIF replays; airline operations on real data (delay causes, airport good and bad days, cancellations, storm cancellations, probability recalibration, passenger rebooking, ferried spares, a live status feed and recovery and reserve planning in the dashboard); airline operations (day-ahead and live forecasts, recovery search, reserve and buffer planning, calibration and backtesting); YAML flowchart models; side-by-side replay; fifteen built-in models with what-if presets; Pareto search; input-uncertainty and rare-event analysis; the cross-entropy method; an in-browser demo; model plugins through entry points; validation of the built-in models against theory.

Out of scope

  • A graphical drag-and-drop model editor. Flowchart YAML plus the dashboard covers simple models; anything more is clearer in Python.
  • Real-time or hardware-in-the-loop simulation.
  • Domain data sets. All models ship with synthetic data only.