Skip to content

Example gallery

Two runnable, CI-tested, end-to-end examples that exercise every parallelism tier the engine ships. Each is built entirely on the public API — no engine edits — and doubles as the final API dogfooding pass before the 1.0 freeze.

Start with Which parallelism do I need?, then see the tier in action:

Example Domain Tiers shown
Ride-hailing fleet Autonomous robotaxi dispatch over a zone graph Replications (5.2a), PDES sharding (5.2b)
LLM agentic workflow Multi-agent task pipeline over a modelled LLM server Replications (5.4a), strict-mode offload (5.4b)

What each tier maps to

Guarantees every example holds

  • Same-seed determinism — two runs with the same master seed produce identical KPIs.
  • Backend reproducibility — study results are bit-identical across the thread, interpreter, and process backends for a fixed master seed.
  • Trace-equivalence — the sharded run equals its sequential reference bitwise; the offloaded policy equals the inline reference bitwise (strict mode).
  • Stdlib-only, no network — the examples add no runtime dependency, and the agentic example is proven by test to make no LLM or network call.

Regenerate the showcase curves with python scripts/generate_example_charts.py.