Wrenfield Foods (a fictional composite) has signed a private-label contract that its Danville, Illinois plant cannot fill on today's performance, and the owner's investment committee will decide on a $12.4 million fifth line in June. Supervisors say OEE is 79.7%, the corporate dashboard says 63.4%, and the written definition the learner builds says 58.4%. Across six modules the learner computes OEE the plant trusts and attributes its losses, predicts changeover times and each purchase order's chance of shipping by its requested date, and checks their coverage and calibration, builds and verifies a supplied four-order, two-line finite-grid Python scheduler, then analyzes conceptual OR-Tools CP-SAT models with allergen sanitation under 21 CFR 117.135 and lockout time under OSHA 29 CFR 1910.147 as hard constraints, puts a read-only rescheduling agent and number-checked shift briefings around the plan, staffs Saturday crews by the contract's rotation without turning OEE into a people score, and proves the result like for like: a replay is a ceiling, a before-and-after mixes causes, and capacity is not money until a cost line changes. The course ends with a go/no-go package for the committee.
Tools are compared the way plants choose them in October 2026: monitoring subscriptions versus MES modules versus an in-house pipeline; an in-house solver versus Siemens Opcenter APS, SAP PP/DS and sequencing or job scheduling in Dynamics 365; vendor agents versus agents built on model APIs with allow-listed tools and approval before dispatch. Vendors are named to show categories, not as endorsements, and no vendor performance claim is taught as fact. Every module ends in a Decision Lab with business metrics, a cliffhanger and several endings. All data are synthetic, and the course is educational, not legal, food-safety or engineering advice. OT data paths and predictive maintenance are in mfg-210, demand planning in mob-210, quality systems in mfg-235 and pharmaceutical packaging in mfg-270.
Executable build scope: builder/exact_example.py and its complete fixture, expected output and rejection tests use Python standard library only. The larger Wrenfield statistical, CP-SAT and agent outputs are synthetic reading/diagnosis illustrations; their original runtime datasets are not supplied and reproduction is not promised.