Simulation with NitroBot Scenarios

Participants act through their bots. A preview is sealed by its exact hash before submission, and each participant sees only their authorized projection. Conditional publication protects a single official scenario result rather than letting each bot invent its own world state.

The browser supports setup, observation, history and forks. It is not a gameplay cockpit: participants use tools to propose and submit their actions, and durable execution can progress without an open browser.

Scenarios shares pure business-engine code while owning separate persistence, credentials and fictional effects. Its virtual clock describes simulated events; wall-clock worker timing does not grant authority to change operational records.

Forks start from a committed checkpoint with new participant bindings and budget. They support exploring another course of action without rewriting the original history. Shared engine code does not mean shared databases or permission to apply a simulation result to live operations.

The hosted participant runner reserves budget before a paid attempt, validates model output and tracks provider cost evidence. Uncertain dispatch does not authorize another paid call. Settling a charge cannot invent a missing response.

These are execution safeguards, not a promise that a model makes good business decisions. Recorded-response tests exercise the runner without spending money.

The job-shop-1 company model is in source with the frozen job-shop-1.0.0 resolver bundle. It checks proposed machine and qualified-operator assignments against calendars and capacity, consumes a batch’s bill of materials, carries work in progress between operations and creates finished stock only after the routing completes. OEM orders use sealed price quotes; simulated shipments settle the accepted amount and any one-time late penalty.

The job-shop pack supplies example machines, skills, materials, prices and a cut, mill and inspect routing. Riverside Fabrication is a fictional three-company, six-week scenario. Its offline replay fixture exercises three turns through draft, publication, submission and resolution, then checks exact replay hashes with scripted rivals and zero model spend.

This proves the tested simulation behavior. The manufacturing operational command front door is still planned; the fixture does not model purchasing, payroll, tax, scrap or maintenance, and does not perform real factory work.