ResponseCompass: Context-Aware Responder Dispatch

When equipment fails on a factory floor, the closest available technician is often not the best person to send. ResponseCompass is a context-aware dispatch agent inside a persistent, simulated control room. It finds who should respond to each issue right now, and explains why. Application code establishes the evidence: the simulation clock, each responder's location, shift and workload, eligibility, past resolution history, and two transparent baselines (nearest responder and a naive score). Compass, running on AMD Token Factory (MiMo-V2.6-Flash by default), then ranks only the eligible candidates it is given and writes a qualitative rationale. A second model acts as an event planner that proposes realistic incidents. Every model output is validated; if it is slow, rate-limited or invalid, the system falls back to a deterministic path, so dispatch never stops. Offers go out automatically. Responders accept or reject asynchronously, complex jobs pull in a second responder, and central command can override at any time. Each decision keeps a full audit trail, and the event feed flags where Compass disagreed with the baselines. Each responder's true skill is hidden simulation truth that no model ever sees, so outcomes reflect real fit rather than anything the model was told. Operators get a live floor map, issue and responder views, and day, week and month analytics on resolution time, success and downtime. Provider controls (rate limiting, 429 retries, a circuit breaker) keep usage within AMD's limits, and an API-free Offline mode makes the demo reproducible. Built with Node.js, plain JavaScript and a 270+ test suite including Chromium end-to-end tests.