Example domain models

Six runnable models, each a different industry, that together exercise every one of the 12 block types and every advanced feature. They live in this repo's examples/ folder — run any of them from a clone with the CLI:

npx tsx packages/engine/src/cli.ts run examples/restaurant.json --pretty
# or: airport.json · hospital.json · bank.json · factory-line.json · network.json

An engine test builds and runs every example and asserts the set covers all 12 node types, so these stay valid as the engine evolves.

What each model shows

Feature	Airport	Clinic	Bank	Factory	Restaurant
source / queue / resource / sink	●	●	●	●	●
branch routing		by-attribute	shortest-queue	probability
delay (no contention)		●			● (pay bill)
seize / release + pools		●
assign attributes / priority		●	●	●
batch / separate				batch	batch + separate
match (assembly)				●
queue reneging		●	●
queue capacity / balking					●
priority discipline		●	●	●
resource preemption				●
resource failures				●

Every node type appears at least once; an engine test asserts this stays true.

---

✈️ Airport check-in — the textbook M/M/c

source → queue → resource (desks) → sink

The simplest model and the right place to start: a single stream of passengers, one waiting line, and a bank of identical check-in desks. This is a classic M/M/c queue — change the number of servers and watch utilization and average wait trade off. The optimizer can find the cheapest desk count that keeps the wait under a target.

Shows: the four essentials, multi-server resources, utilization vs. wait.

🏥 Emergency clinic — shared beds held across care

arrivals → registration queue (reneging) → registration desk → triage (assign acuity) → branch (by-attribute) → {urgent | routine} → seize(beds) → treat (delay) → release → discharged

A bed is held across the whole episode of care, not just one step — that's a resource pool seized at admission and released at discharge, shared between the urgent and routine paths (urgent seizes at higher priority). Patients renege from the registration line if they wait too long (left-without-being-seen), and a by-attribute branch routes by the acuity stamped at triage.

Shows: resource pools, multi-step holds, reneging (Erlang-A), class routing, priority seizing. (~87 discharged/run, beds ~73% utilized.)

🏦 Bank branch — pick the shortest line

customers → assign (VIP / regular) → branch (shortest-queue) → 3 lines [queue (priority + reneging) → teller] → served

Each customer joins the shortest of three teller lines (join-shortest-queue, measuring queue + busy server), VIPs are stamped with a higher priority so they jump each line, and impatient customers renege if the wait drags on. Compare shortest-queue routing against splitting customers randomly — the balanced policy keeps every line short.

Shows: join-shortest-queue routing, priority discipline, abandonment. (~407 served/run, ~17 leave impatient.)

🏭 Factory line — assemble, machine, palletize

{bodies | lids} → assign (part) → match (assemble) → assign (express/standard) → queue → CNC mill (preemption + breakdowns) → QC branch (probability) → {palletize (batch) → shipped | scrapped}

Two part streams are matched into one assembled product (one body + one lid). The CNC mill breaks down (uptime/repair, with reported availability) and express orders preempt standard ones mid-cut (resume). Good units are batched onto pallets of six before shipping; a probabilistic quality check scraps the rest.

Shows: assembly (match), preemptive priority, machine breakdowns/availability, batching, yield routing. (~264 assembled, mill ~88% available, ~34 preemptions, 8% scrapped.)

🍽️ Restaurant — seat parties, limited waiting area

diners → batch (party of 3) → waiting area (capacity → balking) → tables → pay (delay) → separate (party leaves) → departed

Individual diners are batched into parties that are seated together. The waiting area has a capacity — when it's full, arriving parties balk (turn away). A party holds a table for the whole meal, settles up (a pure delay), then separates back into individuals on the way out.

Shows: batching groups, balking on a full buffer, splitting a temporary batch, pure delay. (at a busy load, tables ~100% utilized and ~36 parties/run turned away.)

---

See the Blocks reference for every parameter, and the tutorial to build your own from scratch.