Actor Model Data Simulation

Seed simulations with a small cast of archetypal actors rather than random ones: a Producer that emits messages at a configurable Poisson rate, a Worker pool behind a Router actor, a FlakyService that crashes with probability p on each message, and a SlowConsumer whose processing time is sampled from a log-normal distribution. This handful reproduces the interesting phenomena — backpressure, restart storms, head-of-line blocking, supervisor escalation — without needing a large synthetic graph. Parameterize rates, crash probability, and mailbox bound as sliders so users can drive the system into each regime live.

Drive the simulation from a single logical clock with deterministic seeding. Every actor action (send, receive, crash, restart) is scheduled on a priority queue keyed by virtual time, and the RNG is seeded per-run so a given seed reproduces the exact message interleaving. This is what makes actor demos trustworthy — users who pause on a crash can step forward one message at a time and see the supervisor's decision tree (:restart / :stop / :escalate) apply deterministically. Emit events to the same log the visualization consumes so sim and UI share one source of truth.

For the chat-REPL variant, treat each REPL session as an actor with its own mailbox and let users type , , , and as commands. Back the REPL with the same simulation engine; the REPL is just another event producer. Pre-canned scripts (, , ) should seed the RNG and replay a known scenario so onboarding users hit the interesting failure modes within the first minute.