Finite-capacity vs infinite-capacity scheduling.

Every ERP produces dates. An order lands with a due date of August 14; the routing says the work takes two weeks; so the system starts it July 31 and calls it planned. That subtraction is the entire scheduling method inside most MRP runs — it's called lead-time offsetting, and it rests on one assumption so old it's become invisible: infinite capacity. Understanding the difference between finite-capacity and infinite-capacity scheduling is understanding why your dates slip while your utilization report says everything is fine.

Infinite capacity doesn't mean anyone believes it

No planner thinks the plant has unlimited machines. "Infinite capacity" means something quieter: the model never checks. The lead time is a fixed parameter typed into a master-data field — two weeks, because it was two weeks once — and the queue time buried inside it is assumed constant no matter how much work lands in the same window. Load doesn't lengthen it. Contention doesn't exist. Forty jobs can be offset through a work center that fits twelve a week, and every one of them gets a start date.

Each date is individually plausible. Collectively they are impossible. The system isn't lying, exactly — it's doing arithmetic with a calendar and calling it a plan.

A plan that cannot say no is not a plan. It's a wish, formatted as a date.

What the fiction costs

• Sales promises what the floor can't sequence. The quoted date came from subtraction, not from a slot on a machine — so the salesperson promises it, the plant misses it, and the plant eats the blame.
• Inventory becomes insurance. Nobody trusts the dates, so everyone buffers against them — safety stock as an apology for a schedule no one believes. Working capital freezes where confidence should be.
• Expediting becomes the real system. The hot list overrides the plan daily, which teaches everyone the plan is decorative — which makes the hot list longer.
• A planner becomes the actual scheduler. The single most consequential planning layer in the company ends up in a spreadsheet, resequenced by hand every morning from memory and gut.

What finite-capacity scheduling actually models

Finite-capacity scheduling starts from the opposite premise: a date is only real if it corresponds to a feasible slot. That requires modeling the things lead-time offsetting ignores:

• A real calendar. Shifts, holidays, maintenance windows — time as the floor experiences it, not a 24/7 abstraction.
• Contention. Jobs queue because the model makes them queue. Queue time is a consequence of load, not a typed-in constant.
• Sequence-dependent changeovers. What follows what matters — changeover families, allergen and cleaning boundaries, the make-ready that eats a shift when the sequence is wrong.
• Eligibility as a bundle. A batch doesn't need "a machine." It needs the press, the tooling, and a qualified room — free at the same time. Checking one resource and hoping for the rest is how schedules lie politely.
• Materials, pegged. The start date holds only if the inputs exist — so supply is netted and pegged to the demand it serves, not assumed.

Model all of that and something changes philosophically: the system now knows the feasible region — the set of schedules your constraints actually permit. It can finally say no. Which is the only reason its yes means anything: the most efficient plan your constraints permit, found continuously — and the feasible region is known exactly, because it was modeled.

If finite is obviously right, why is infinite still everywhere?

Because finite-capacity scheduling needs a model of your factory, and the industry's answer to that has been brutal: author it by hand. The classic APS deployment sends consultants to interview your planners and type your routings, changeover matrices, and resource rules into a modeling studio for six to eighteen months — enterprise-priced, and stale the day it goes live because the floor moved on during the project. The big suites hand off at the factory door; the mid-market — most of manufacturing — was priced out of the fix entirely. So the fiction survived, not because anyone defends it, but because the cure cost more than the disease.

The model you don't build

That authoring project is the part that shouldn't exist. Your industry already knows its own physics — a pharma plant knows its changeover families and validated boundaries, a converter knows its make-readies, a shrimp processor knows its grade-count yield splits. PatternLab Cortex starts with that knowledge encoded as a ready-made industry model. AI-assisted mapping connects existing fields, records and documents to the model and resolves identities across inconsistent names and codes; every mapping remains reviewable, traceable and versioned. A deterministic finite-capacity solver then evaluates schedules inside the feasible region. AI proposes. Rules and solvers validate. People approve high-impact decisions. Your ERP stays the system of record; Cortex is the operational semantic layer, and its live Supply Chain workspace connects Plan, Inventory and Sales above it.

Where infinite capacity is honest

Fairness requires saying so: at the aggregate, long-horizon altitude — annual capacity strategy, rough-cut S&OP, the IBP layer — infinite-capacity math is a reasonable simplification, because nobody is promising a customer a Tuesday. The dishonesty begins below the handoff, where "what runs next, on which line, promised for when" gets decided. That layer — the one the suites don't serve — is where finite capacity stops being a nice-to-have and becomes the difference between a commitment and a guess.

Frequently asked

Is MRP finite-capacity or infinite-capacity?

Infinite. MRP nets material requirements and offsets fixed lead times against due dates; it never checks whether the work centers those lead times imply are actually free. That's why MRP dates drift the moment the plant is busy — the busier the floor, the more fictional the fixed queue time inside the lead time becomes.

What's the difference between APS and MRP scheduling?

MRP answers "what materials, roughly when" with infinite-capacity arithmetic. APS (advanced planning & scheduling) answers "what runs next, on which resource" against finite capacity — calendars, contention and changeovers. The historical catch is the work required to author and maintain a factory model. PatternLab Cortex starts from a ready-made industry model and activates it through reviewable AI-assisted mapping and identity resolution.

Do I need finite-capacity scheduling if I already have an ERP?

The ERP stays your system of record either way. If quoted dates slip while utilization reports look fine, add a finite-capacity planning layer that connects ERP orders to real materials, calendars and constraints rather than replacing the ERP.