Work in progress. This showcase is under active development and changes as the engine is tested. The companies, their data and the disruptions are fictional; every number is what the engine computed on them.
Sites at their real geographic coordinates. Solid lines are routes used in the baseline plan; dashed lines are available backup or co-packer routes. Highlighted nodes show where the active disruption story moves production or inventory.
Who decides: the S&OP release gate
Human-in-the-loop governance
Promise an order (Available to Promise)
Telescoping daily / weekly / monthly horizon
The AI explains; it never approves
Deterministic Master-Data Contract Check
Forecasting, used only when it has earned it
Out-of-sample rolling backtest
Each model forecasts weeks it never saw, and is scored on what really happened. Classic statistical models and machine-learning models compete; the plan admits only a model that beats the baseline on held-out error.
Scaling & engineering roadmap
Measured benchmarks on 2 CPUs
Plans, not claims: what is still open, how far each item is, and what was measured.