rule 150

Mapping begins with the customer requirements; no customer node is present.

The door-to-door map must include delivery of at least one main raw material.

Information flow is missing. In lean manufacturing it is treated with as much importance as material flow — you must map both.

Draw a timeline under the process boxes and inventory triangles to compile production lead time.

A data box is drawn under each process box. Processes are missing their data.

One process box equals an area of flow, and all processes should be labelled.

Count and time should both be noted below every inventory triangle.

A FIFO lane limits quantity — the maximum quantity must be indicated.

"Buffer" or "Safety Stock" must be noted on the closed-sided store icon.

Note the frequency of shipments on every truck icon.

The lower rung of the timeline does not match the cycle times in the data boxes above it. The same figure is drawn twice on a value stream map, and the two disagree, so one of them is wrong. Re-time the process and redraw the ladder from what the stopwatch says.

The value-adding time quoted at the foot of the map is not the sum of the cycle times in the data boxes. That total is the number the whole map argues about — the fraction of lead time spent adding value — so it cannot be carried separately from the figures it comes from.

Material flow is drawn left to right on the bottom half of the map, in processing order — not by plant layout.

Information flow is drawn right to left across the top half of the map.

Value stream mapping means walking one product family from door to door. Drawing all product flows on one map is too complicated.

On the future-state map takt times are noted in the data boxes. Question 1 has not been answered.

Subtract breaks, meetings and cleanup from shift length before dividing by customer demand.

The pacemaker cycles slower than takt and cannot meet customer demand within available time.

Cycling much faster than takt compensates for production problems and removes the incentive to eliminate them. If you cycle faster than takt there should be a plan for closing the gap.

Cycling below takt is intentional here: it leaves time in the shift for the required changeovers.

Continuous flow gives the biggest reduction in waste and lead time. Do not set up a pull system if you can create continuous flow.

Adjacent dedicated processes have cycle times close to each other and to takt — they can be placed in a continuous flow.

Continuous flow means each item passes immediately to the next step, with at most one piece at automated processes.

Incorporating a fast, shared process into a continuous flow would slow it to takt and dedicate it to one family — a vastly underutilised machine. Control it with a supermarket pull instead.

Work content divided by takt leaves a fractional operator. Reduce work content by process kaizen to bring it under the takt ceiling, or plan overtime.

Distribute work elements so that each operator's work content is just below takt time.

Coupling processes merges their down times: with no stock at the boundary, the cell stops whenever any member of it stops. One or more coupled processes here do not hold the availability their own load and recovery require — not a general reliability standard, but the figure that this demand, this effective cycle and this mean time to recover imply for this boundary. Either raise availability to that figure, take work content out to create the margin, or put a FIFO lane at the boundary long enough to cover the recovery.

Push means producing regardless of the downstream need. A lean future state links every process to its customer by continuous flow or pull.

Resist scheduling batch processes independently — a schedule is only an estimate of what the next process will need. Link them to their downstream customer via pull.

Withdrawals trigger kanban movement from the supermarket to the supplying process, where they are the only production instruction.

To qualify as pull, parts must not be produced or conveyed when there is no kanban, and quantity must match the kanban. "Go see" adjustments are not a true pull.

With very many configurations it is impractical to stock all variations. Use a FIFO lane between decoupled processes instead of a supermarket.

Replenishing bin-for-bin is impractical when a changeover costs more than a container of run time. Use a signal kanban at a trigger point to launch a predetermined batch.

Supermarkets should be located near the supplying process so it keeps a visual sense of customer usage.

If the FIFO lane gets full the supplying process must stop producing until the customer has used some inventory.

Sequenced pull works only if lead time in the supplying process is short enough for production to order.

Safety stock should be temporary — used only until the root cause is found and eliminated. There should be strict rules for using it.

Send the customer schedule to only one production process — the pacemaker. This single point regulates the entire value stream.

All material transfers downstream of the pacemaker process need to occur as a flow — no supermarkets or pulls.

The pacemaker is where the customer's order rate enters the stream, so it belongs at the last point that gives the product its form. Processes downstream of it should flow to its pace without a schedule of their own — and here one or more operations downstream are still transforming the product, which leaves that stretch of the stream taking instructions from somewhere other than the pace. A control point downstream is not this defect: an inspection confirms the form rather than conferring it, and it can flow FIFO on the pacemaker's pace. High product variety is a reason to move the pacemaker upstream of a build-to-order area — one step upstream, not to the head of the stream.

With custom products and job shops the scheduling point often needs to be further upstream, where product variation arises, with FIFO flow downstream of it.

Get rid of those elements of MRP that try to schedule different areas of the plant. Let downstream withdrawals determine upstream production.

Adjusting schedules by checking inventory levels is not a true pull. Incorporate it honestly on the current state, then design it out.

Your order-to-delivery lead time involves only those processes from the pacemaker downstream to delivery.

Building directly to shipping requires either a reliable short-lead-time order-to-delivery stream, or more safety stock.

With multiple products, lack of levelling means you are still batching your builds. Level the mix at the pacemaker.

Even if you produce only one product you still need to level the production volume.

Question 7 is unanswered: state the consistent increment of work released and taken away at the pacemaker.

Pitch is takt time multiplied up to a finished-goods transfer quantity: pitch = takt × pack size.

Takt here is longer than a whole shift, so it cannot be used to pace work at the station. Treat it as the capacity check it is, and manage the stream by the schedule and the queue in front of the constraint.

Release only a small consistent amount of production instruction — usually between 5 and 60 minutes worth — at the pacemaker.

Returning a whole batch of kanban to the cell provides no takt image. Bring one kanban at a time at the pitch increment.

To allow production to takt and mix levelling, the pacemaker should incur little or no changeover time and change over very frequently.

Frequent changeovers require keeping all fixtures and component varieties near the operator's fingertips — and failsafe devices to prevent the wrong ones being assembled.

Leveling means alternating repeatedly between smaller batches, not producing all of one type then all of another.

Note either the batch sizes or EPE in the data boxes of fabrication processes upstream of the pacemaker.

An initial goal at many plants is to make at least every part every day for high-running part numbers.

A typical target is approximately 10% of available time used for changeovers. Reduce changeover time or improve uptime to run smaller batches.

The declared batch size requires more changeovers per day than the available time allows.

Available work time divided by cycle time multiplied by uptime is below the demand this process must serve.

Production lead time cannot be shorter than the sum of value-adding times.

Lead time for each inventory triangle is inventory quantity divided by the daily customer requirement.

Total production lead time is the sum of the lead times through each process and each inventory triangle.

For maps with multiple upstream flows, use the longest time path to compute total lead time.

A shorter production lead time increases inventory turns; the two figures must be consistent.

Comparing value-added time to total lead time should give you quite a shock — this is the core insight the map exists to deliver.

Size the supermarket to cover the replenishment delay, with a modest extra allowance for supplying-process problems.

This map calls itself ready to implement while facts it depends on were assumed rather than observed. Walk the floor and establish them, or call the map a hypothesis.

A future state becomes a commitment when the functions that have to deliver it have agreed it. Record the sign-offs before implementing against this map.

The working days a month and the working days a year on this map do not multiply out. Every rate derived from them — takt, pitch, the annual plan, the benefit case — inherits the difference.

Where flow starts and stops rests in part on inputs nobody supplied, so those boundaries are drawn the safe way and marked provisional. State changeover frequency, the families and SKUs on each shared resource, and demand by family: the answer may allow a tighter link than the one drawn here, and until it arrives this stretch of the design is a proposal rather than a plan.

The ideal state has been given a date. It is the condition to steer by once every enabling condition has landed; dating it turns a direction into a promise nobody made.

A buffer left on the future state without a stated cause and a condition that would remove it becomes permanent by default. Name what each one is absorbing.

A plan item gives its kaizen event the same duration as the change it lands. The event runs three to five days; standard work, training and holding the result take longer, and reporting one number for both understates the programme.

A defect family is attributed to the station that detected it. A countermeasure aimed there can only inspect harder; find the operation that creates the defect and prevent it at source.

A future state that does not shorten production lead time has not eliminated any source of waste.

Question 8 is unanswered: mark the equipment and procedural improvements the design depends on with kaizen lightning bursts.

One person needs lead responsibility for understanding and improving the product family's value stream, reporting to the top person on site.

Do not split the mapping task among area managers and stitch their segments together. Map the whole value stream yourself.

Always collect current-state information while walking the actual flows yourself. Numbers in a file rarely reflect current reality.

Begin at the shipping end and work upstream, so you start with the processes linked most directly to the customer.

A current-state map and the effort to create it are pure muda unless you use it to create and implement a future-state map.

As your future state becomes reality, a new future-state map should be drawn. There must always be a future-state map.

The yearly value stream plan must state exactly what will be done by when, with measurable goals and clear checkpoints with real deadlines and named reviewers.

A future-state map with no value stream plan behind it changes nothing. Split the future state into loops and commit each one to a named owner with a deadline, or the drawing will hang on the wall while the stream stays as it is.

One effective strategy is to begin implementation in the downstream pacemaker loop and steadily transition upstream.

Improvements to a loop follow a pattern: develop continuous flow at takt, establish pull, introduce levelling, then practise kaizen continuously.

When in doubt, move forward on flow improvements and let them drive the supporting process improvements. Do not get stuck in process improvement.

Kick off improvement projects by creating a pull: state that in 30 days the batch size will be reduced, rather than pushing a team to reduce setup time.

Measures emphasising equipment utilisation or overhead absorption have a negative impact on flow. Measure lead time, space, effort, defects and missed deliveries instead.

If your support organisation cannot respond to operator problems within takt time, the organisation is not ready for production to operate at takt.

Constant efforts to eliminate overproduction are what create great flow. This future state contains no countermeasure against it.

Total work content divided by takt requires more operators than the map staffs. No assignment of the work to this few people can meet the schedule — the shortfall is in the arithmetic, not in the balance.

A cell cycles faster than the work content assigned to each of its operators, which cannot happen: the effective cycle time is the largest of the operator times, so it never falls below their mean. Either the cell needs more operators or its work content is overstated.

A fifth or more of a cell's paid operator time is standing idle because the work is unevenly assigned. Redistribute work elements towards the average operator time before adding operators or equipment — the capacity is already bought.

A station whose work is one indivisible task longer than takt sets the pace of the whole line on its own, and no rebalancing can change that. Run the station in parallel so alternate units pass through each copy.

Parallel stations come in whole copies, and enough of them to bring the station under takt: k copies of a task of length t deliver a unit every t/k, so k must be at least the ratio of the two.

The map has more producing stations than operators to staff them. A station may be worked by two people, but not by none.

There are fewer work elements than operators can be balanced across. Break the tasks down further — the balance can only ever be as even as the largest element allows.

These variants share less than a third of their work content, so balancing one line for all of them wastes most of it. Either map them separately or confirm they really pass through the same processes.

More than one part in twenty fails somewhere across the stream, and every one of them was paid for at each process it passed. Attack the losing process at its own station rather than sorting at the end — a defect caught where it is made costs one process, a defect found at inspection costs all of them.

On uptime alone this stream covers its demand; once the stated speed and quality losses are applied it does not. Capacity has to be read on good parts out, not running time — the difference is being paid in overtime or in a shortfall someone absorbs, and it will not appear in any single data box on this map.

A process in this future state cannot cover demand once its own stated speed and quality losses are applied, and nothing on the map says what closes the gap. A future state is a commitment to run this way, so the shortfall has to be answered where it is drawn: move work content off the short process, add the station, or recover the effectiveness losses and show the figure you are recovering to. Left unnamed it becomes overtime discovered in the third week.

A process drawn inside a continuous-flow cell is available less than half the time. Its load may be light enough that the capacity arithmetic clears it, but that arithmetic is answering the wrong question: with no stock between the processes, everything after it stops for the whole of every repair. Take it out of the cell behind a FIFO lane, or raise its availability before the boundary is designed at all.

A cell on this map cannot hold its pace even though every process in it passes on its own. Coupling multiplies availability rather than averaging it — with no stock between them, a stoppage anywhere stops everything, so three processes at 90% run as one at 73%. Either raise the members above what the cell as a whole requires, or split the cell at its weakest boundary and let a FIFO lane absorb the stoppage.

The scheduling point sits on a station that verifies or moves the product rather than one that gives it its form. Everything that actually transforms the product is upstream of it, so those processes will each need a schedule of their own and the stream ends up with several paces instead of one. Put the pacemaker at the last process that confers form, and let the checks and the packing flow FIFO on its pace.

The pacemaker was close to being placed elsewhere, yet the map shows only the winner. When the margin is thin the runner-up matters: the team knows things the arithmetic does not, and this is the decision they are best placed to overturn. Publish the candidates, the criteria they were scored against, and the margin between the first two.

A process is treated as a monument without anything on the map establishing it as one. The label is doing real work here — it excuses the batch and the inventory on both sides — so it needs grounds: a minimum load the equipment cannot go below, a physical or thermal constraint, or another family with a claim on the same machine. A long changeover on its own is an improvement opportunity, not a monument.

A batch process runs on an interval longer than the stock placed after it can cover. The buffer has to hold the demand between loads *and* the pieces the next process would consume while a failure mid-load was being recovered; sized for the interval alone it empties the first time the load is late. Either raise the buffer to cover interval plus recovery, or run the process more often on a smaller load.

Improvement work on this map names the process and the theme but not the condition it is meant to create. Without a target condition there is no point at which the work is finished, and the boundary that depends on it stays unsecured while the item is reported as in progress. State where the measure stands today, where it has to reach for the design to hold, and how anyone would see it got there.

Processes have been coupled into continuous flow without the map saying what made that possible. Coupling merges the downtime, changeovers and yield losses of both into a single system, so the assessment behind it is the part the team needs: which factors were tested, which one was binding, and what has to stay true for the coupling to hold.

Parts travel a long way between processes that are pushing to each other. Distance is a symptom of the layout, not of the work: bringing the steps into flow order typically removes most of it without changing a single cycle time.

A process on this map transforms a whole charge at once, and its equivalent rate per piece sits under takt — which makes it look like spare capacity. It is not: it produces nothing for a whole batch cycle and then a batch at once. The equivalent rate is a batch cycle divided by the pieces the batch yields, and it describes no moment of the operation, so read the batch figures beside it before drawing any conclusion about headroom here.

A process here has been treated as a batch operation because of what it is called, with no batch size, working volume or residence time recorded anywhere on the box. That may well be right, and nothing downstream of it can be sized until it is confirmed: the charge decides the buffer, the interval and whether the process can be coupled at all.

A batch process on this map is missing figures its own arithmetic needs — batch size, working volume, residence time, cleaning time or yield. Capacity, the interval it forces downstream and the buffer that covers it all follow from those, so until they are recorded the design around this process rests on a per-piece average rather than on the operation.

A process that transforms a whole charge has been placed inside a continuous-flow cell. There is nothing for the next operation to pick up until the charge is finished, so the cell stands idle for a batch cycle and then receives a batch at once — which is not flow, whatever the layout looks like. A moving-conveyor process can join a cell because a piece can be set down and one picked up within a takt; a vessel cannot. Decouple it with a FIFO lane or a hold tank and pace the cell from somewhere else.

A continuous-flow cell here contains a process whose name suggests it transforms a whole charge, and nothing on the box settles the question. If it does, the cell will stand idle for a batch cycle and then receive a batch at once, so the coupling fails rather than underperforms. Establish whether pieces pass through one at a time — a moving conveyor through an oven does and belongs in the cell; a vessel does not.

A data box on this map claims an every-part-every-day interval that the equipment cannot hold. Cycling it through the grade families that share it, with the clean-down each transition costs, takes longer than the interval claimed. Every part every day is the right goal and it applies to process industries too, but it is reached by cutting the clean-down, not by writing the target into the box: state the interval the equipment can hold today, and the reduction that would close the gap.

The interval stated for a shared piece of equipment is shorter than the time it takes to run one campaign of every grade that uses it. The interval is the sum of the processing time for each grade at its batch size, plus the clean-down between them, over the working time available — so a shorter figure is not a stretch target, it is arithmetic that does not close, and any supermarket sized from it will run dry.

A batch process here has no cleaning time recorded. For a vessel or a mill the clean-down is the changeover, and it is sequence-dependent: running light to dark needs a flush, running dark to light needs the vessel stripped. Without it the cost of a grade change is unknown, so the campaign length, the sequence and the interval are all unfounded.

Once the charge, the residence time, the yield and the availability of a batch process here are put together, it does not cover the demand placed on it. This is not a boundary problem and no buffer solves it: the equipment cannot make the volume, so either the batch has to grow, the cycle has to shorten, or the requirement has to be met somewhere else.

A buffer on this map holds fewer pieces than the process filling it delivers in one charge. That process discharges a whole batch or nothing, so the first replenishment overflows the location and the pull signal in front of it asks for a fraction of a batch that cannot be run. Size the location to a whole number of charges, or cut the charge until the size you want is a multiple of it.

A batch process on this map is released by laboratory test, and how long the test takes is not recorded. That wait is lead time and it is stock: the charge is made, it is in the building, and nobody may ship it. Until the hold is stated the lead time is understated by however long it runs, and any supermarket fed by this process is short by the quantity sitting under test.

A batch process here does not say how its output becomes shippable stock. If it releases on a parameter read off the vessel, the charge is available at discharge; if it releases on a laboratory certificate, it is available whenever the laboratory finishes. Those are different streams with different lead times and different buffers, and the map currently describes neither.

Most of the transformations on this stream are charges rather than pieces, and the process types have been read off the process names rather than stated. Which arithmetic governs each operation — per piece, per batch or per unit of rate — decides the capacity, the boundaries and every buffer size on the map, so it is the first thing to confirm on the floor rather than a detail to settle later.

A cell on this map is drawn with more operator work content than its operators can perform inside the takt time, and adding the operator the remainder rule would add does not close the gap. No rebalancing fixes this: the work content itself exceeds the time available. Either work comes out of the cell or the cell does not make the volume.

A cell on this map is staffed at a number that does not follow from its work content. Operators are total operator work content divided by takt time, and the remainder decides whether to add one: below 0.3 do not, between 0.3 and 0.5 do not yet, above 0.5 add one and keep working to remove the need. Where the count is higher than that, the usual reason is that machine time has been charged to the operator — an operator who has loaded a machine and started it is free while it cycles, and staffing for the wait builds the waiting in.

A cell on this map contains operations of more than one kind but is presented as a single category, or does not record what it contains at all. A cell that machines, washes, inspects and packs is a mixed production cell, and each operation has to keep its own type: a defect found in the cell is traced back to the operation that made it, and a cell reduced to one label cannot be traced through.

A FIFO lane on this map holds less than one mean recovery of the process feeding it. That is the one thing a lane is for: when the process upstream stops, the lane is what keeps the process downstream running until it comes back. Sized below the recovery it empties first, and the boundary is decoupled on the drawing and coupled in the building.

At a boundary on this map, covering one recovery of the feeding process needs more pieces than the process downstream consumes in a whole replenishment interval. No lane maximum satisfies both — large enough to protect the boundary is larger than the boundary turns over. Either the recovery comes down or this boundary needs a supermarket rather than a lane.

A process on this map claims to run every part more often than its changeover time allows. The time left for changeovers is the available time less the time to run the requirement, and that divided by the changeover time is how many changeovers fit in a day. Where the claimed interval needs more changeovers than fit, the interval is a target rather than a plan, and it needs the setup reduction beside it that would make it true.

A process on this map cannot make the daily volume once its own losses are taken out. A cycle time inside takt is not capacity: capacity is the available time divided by the cycle time, times uptime, less the time spent changing over. No buffer fixes a shortfall here — a buffer only delays the day the shortfall becomes visible, and it is refilled by the process that cannot fill it.

A machine inside a proposed cell has an effective cycle time above 80% of takt, counting its own cycle, the load and unload it cannot cycle during, and its changeover amortised over the run. Operators will come back round to it before it has finished and wait — which is the waste the cell was built to remove, reintroduced by the equipment. Either that machine gets faster or the cell needs a buffer at it.

A cell on this map is designed without an observed walking time, because the physical layout does not exist yet. That is the right way round — walking is left out while the cell is on paper, precisely because the layout is unknown — but it means the cycle time shown is the paper figure. Lay the cell out, walk it, and time it before this design becomes an instruction. In the worked example in the book the walk came to six seconds on 88 seconds of content, and it was enough to change the staffing conversation.

At a boundary on this map, a mean stoppage upstream costs more pieces than the buffer downstream absorbs, so part of every stoppage reaches the next process. Whether that matters depends on the spare capacity available to make it back: with enough, the exposure is real and short. The figures are on the map so the choice between a longer buffer and a faster repair can be argued rather than assumed.

A FIFO lane on this map is sized without a recorded mean time to recover at the process feeding it. Its quantity is therefore a fraction of a day's demand rather than a derivation — a policy, not a design, and nothing about the boundary would change it. Uptime says how often a process stops; only the recovery says how long for, and only the second sizes a lane.

The capacity figures on this map are derived from the data boxes and have not been checked against what the processes actually produced. Nominal capacity, effective capacity and every buffer sized from them are arithmetic on recorded cycle times and uptimes — all internally consistent, and all untested. A week of counted output at the tightest processes would price every assumption above it at once.

A process on this map is recorded as shared with other product families, but how much of its time those families take is not. The interval at which this stream can withdraw from it depends on everything the machine has to make, not only on this part, so the interval shown is the shortest it could be rather than the one it will be. What is needed is the demand by family on that resource.

More than one item in five has to go back for missing or wrong information somewhere on the stream. That loop-back is lead time nobody's cycle time shows. Fix the incomplete hand-off at the process that creates it — a downstream checker is judgment inspection by another name.

The operator spends more than a quarter of the cycle walking. That is a layout number, not a work-content number: a U-cell or a shorter pitch between machines gives the time back without touching the job.

At one or more stations the machine's running time and the operator's walking together account for less than a tenth of the cycle time recorded on the box. Work content is measured element by element; a station total that its own elements cannot reach was estimated rather than timed, and the usual cause is a figure written in minutes into a field that reads seconds. Every other number on this map has been derived from it.

A cell with several stations and no stated standard WIP will fill with whatever the operators feel safe holding. One-piece flow is a WIP rule, not a layout drawing.

This process spends more than a fifth of the shift changing over. Separate internal from external setup and convert what you can — the EPE and the supermarket both shrink with the changeover, not with a larger batch.

The stream loses parts and names no source-inspection device. Sorting at the end is judgment inspection: it finds the defect after the cost is paid. A poka-yoke at the process that makes the defect is the only inspection that can reach zero.

Every minute of this setup is internal: nothing has been moved off-line. That is SMED's preliminary stage, and stages 1 and 2 — separate internal from external, then convert — routinely take half the stopped time out before anybody buys a clamp.

Internal setup runs past ten minutes, so the changeover is outside the single-minute range SMED aims at. Functional clamps, an intermediate jig and standard die heights are the three that usually get it there, in that order.

The stated changeover is longer than the internal setup, which means it counts staging or preparation the machine is not waiting for. Batch size, EPE and capacity all follow the internal figure; using the total overstates the loss and hides the work that is already external.

A machine here runs longer unattended than its operator takes to load and unload it, and the operator is tied to one machine anyway. That is the case Shingo answers with multi-process handling: give the operator the next process to walk to, and auto-eject so the part comes off without a hand.

Three or more operations sit comfortably below takt, each carrying a little waiting. Rother and Harris are blunt about this: spread evenly, the waste is permanent and invisible. Load all but one operator to nearly a full takt and leave the slack on one, where it can be seen and removed.

Demand is not stable, the stream builds straight to shipping, and no buffer is named. Something absorbs that variation — overtime, expediting, or a late shipment. Tapping and Shuker's answer is an explicit buffer resource, sized and available immediately, rather than an implicit one paid for in expediting.