Pull SimulatorPull System Simulator

What you are watching: consumption as the instruction

Most explanations of a pull system describe the mechanics — cards, bins, supermarkets — without ever showing the thing that makes it remarkable: nobody in this plant is told what to do by a schedule, except at one single point. A customer takes a box from the finished-goods store, and that act of consumption ripples upstream — store to cell, cell to parts market, market to supplier — as a chain of kanban cards, each one released by the consumption of the material it was attached to. Material flows left to right across the floor; information flows right to left. The simulator makes both flows visible at once, which is nearly impossible to do standing on a real shop floor.

Two rules run everything. Nothing moves without a blue withdrawal kanban; nothing is made without a green production kanban. Every store on the floor — the finished-goods supermarket and the purchased-parts market alike — is a place where those two cards swap: the withdrawal card carries material downstream, and the production card it displaces carries the build signal upstream.

The one schedule: a heijunka box

The day begins before the shift does, at the heijunka box. Demand is 240 pieces over 480 minutes — takt of 2 minutes — and boxes hold 20 pieces, so one box must ship every 40 minutes. That interval is the pitch, and it gives the box twelve slots. How those slots are loaded is half the lesson: the product-family ratio of 3:2:1 becomes the interleaved sequence A-B-A-B-A-C, repeating — leveled by volume and mix. Load the same cards as blocks instead and you manufacture your own demand spikes; the experiment mode lets you watch that happen.

Every pitch, a runner takes the next card and pulls exactly that box — no more, no less — from finished goods to the shipping lane. The pitch is a heartbeat for management, too: a problem anywhere in the plant surfaces as an empty store within one pitch, not as a surprise at the end of the month.

From finished goods to raw material, one swap at a time

Follow one card through the tour and the whole system unfolds. The withdrawal kanban rides the box of A to shipping. The production kanban that box was carrying comes off at the store and travels to the cell’s kanban post — the cell’s entire schedule is that queue of green cards, arriving in the leveled sequence the heijunka box created. One operator works the U at takt, one lap per piece. The finished box carries its production kanban back to the store, closing the first loop.

Meanwhile the build emptied two bins at the cell’s flow racks. Their withdrawal cards go out with the water spider on a timed route — standardized work for material handling, the same discipline as the milk run — and at the parts market the identical two-card swap happens one loop upstream: full bin out with the withdrawal card, supplier kanban off to the supplier post. When the supplier truck docks at its fixed times, it delivers one bin per waiting card. The pull chain is complete: finished goods to raw material, with no schedule anywhere but the leveling box.

Then try to break it

The experiment mode exists because the deepest lessons are counterfactual. Four levers:

1. Batch the heijunka box.Same twelve cards, loaded as blocks. Watch each family’s store crash toward zero while its block runs — bullwhip generated entirely inside your own four walls. It survives only because the store is sized generously.
2. Cut the store. One card fewer per family. With leveled loading, nothing happens — the plant runs a third leaner, proving inventory is a design decision. Combine it with batching and the runner starts missing pitches.
3. Spike demand 25%. The pitch tightens to 32 minutes but nobody resized the loops — and the plant tells you within pitches, not weeks. The fix is arithmetic, not heroics: recalculate takt, pitch, and card counts.
4. Stop the customer at noon. The pulls stop, so the cards stop, so the cell, the spider, and the supplier post all go quiet on their own. A push plant would have kept producing to schedule all afternoon. Overproduction is a system property, and pull deletes it by design.

Why WIP stays flat all day

Keep an eye on the WIP counter during the final chapter. It barely moves, because in a kanban loop inventory physically cannot exceed the number of cards in circulation — overproducing would require a free production kanban, and none exist beyond the designed count. That cap is what stabilizes lead time (Little’s Law: lead time = WIP ÷ throughput), and it is why sizing the loop is a calculation rather than a guess. When you are ready to size a real one, the Kanban Card Calculator does the arithmetic this simulation animates, and the PFEP Builder holds the part-by-part data a real market is built from.

Running it as a workshop exercise

The simulator is built to be projected. A sequence that works well in training rooms:

1. Tour first. Walk the chapters and narrate the two-card swaps aloud — the store swap in chapter 4 is the moment to slow down for.
2. Ask the room to find the schedule. Most people look for one at every process. There is exactly one, and finding it lands the point better than stating it.
3. Predict, then run, each experiment.“What happens if we load the box in batches?” Let the room commit to an answer before you press run.
4. Transfer.Ask: “In our plant, what tells each process what to make — a card, or a guess?”

What the simulation assumes

The model is deliberately clean so the card logic stays legible: one cell, two purchased components, fixed cycle times, no changeovers, no variability, and travel times measured in minutes. Real pull systems must also cover changeover batching (see EPEI), demand variation buffered by safety stock, and signal kanban for batch processes — refinements, not exceptions, to the two rules you watched. The tour’s plant is the textbook future state of value stream mapping: supplier loop, supermarket pull, a leveled pacemaker, and pitch withdrawal.

Frequently asked questions

What is a pull system?

A pull system is a way of controlling production and material movement in which nothing is made or moved until downstream consumption signals the need. Instead of pushing material through the plant according to a forecast-driven schedule, each process withdraws from an upstream store (a supermarket), and that withdrawal releases a kanban card authorizing the upstream process to replace exactly what was taken. The customer sets the pace; the signal travels backward from finished goods toward raw material while material flows forward.

How does a two-card kanban system work?

A two-card system uses withdrawal kanban and production kanban. A withdrawal kanban is permission to MOVE: it travels with material from a store to the point of use. A production kanban is permission to MAKE: it stays with a container in the store, and when that container is withdrawn, the production kanban is detached and sent to the producing process as its build instruction. The two cards meet and swap at every store — at finished goods when the runner pulls a box, and again at the parts market when the water spider exchanges an empty bin for a full one.

What is a heijunka box and why is it loaded by ratio?

A heijunka box (leveling box) is a scheduling device with one column per pitch — one timed slot per container of output — and one row per product family. It is loaded according to the family mix ratio, interleaved rather than in blocks: with a ratio of 3:2:1 the sequence is A-B-A-B-A-C repeating, not six A boxes followed by four B boxes. Leveling by volume and mix means every downstream process sees smooth, repeating demand, which is what allows supermarkets and kanban loops to be small.

What is a pitch pull?

Pitch is takt time multiplied by pack-out quantity — the time it takes for demand to consume one container. In this simulation takt is 2 minutes and boxes hold 20 pieces, so the pitch is 40 minutes. Every pitch, a material handler takes the next withdrawal kanban from the heijunka box and pulls exactly that box from the finished-goods store to the shipping lane. The pitch is also a management timeframe: if the runner finds an empty store, the plant knows it is behind within 40 minutes, not at the end of the shift.

Why does the simulation say WIP is "capped by design"?

In a kanban loop, every container in circulation must carry a card, and the number of cards is fixed. Inventory therefore cannot exceed the card count no matter what any individual process decides to do — there is no way to overproduce, because producing requires a free production kanban and none exist beyond the designed number. That is why the WIP counter stays flat all day. In a push system, WIP is whatever the schedules happen to produce, which is why it balloons whenever forecasts and reality diverge.

What happens if you load the heijunka box in batches instead of leveling it?

The experiment mode lets you try exactly this. With batched loading (all A boxes, then all B, then all C), each family’s finished-goods inventory crashes while its block runs and sits idle otherwise — demand amplification created entirely inside the plant. With a generously sized store the plant survives; combine batching with a lean store and the runner starts missing pitches. The lesson: leveling is not cosmetic. It is precisely what allows inventory to be small.

Is this pull system simulation free?

Yes. It runs entirely in your browser with no signup, no install, and no data collection. Take the guided tour, run the experiments, and export a print-ready summary. To size a real loop with your own numbers, the free Kanban Card Calculator, Heijunka Box Designer, and Milk Run Planner cover the calculations this simulation animates.

Related tools

• Kanban Card Calculator — size a real replenishment loop from demand, lead time, and container quantity
• Heijunka Box Designer — build the leveling box this plant runs on, from your own demand and pack-out data
• Milk Run Planner — design the water spider’s timed route with standard times per stop
• PFEP Builder — the part-by-part database a purchased-parts market is sized from
• One-Piece Flow Simulator — the companion simulation: flow within a process, where pull connects processes