Cycle or Chase - why a fixed cycle recovers & re-planning does not

Cycle or Chase

Why a fixed cycle recovers & re-planning does not

When stock runs short, the natural response is to rebuild the schedule around it. Each rebuild costs changeover time & pushes other work back, which sets up next week's shortage. That is how a one-week disruption keeps a plant re-planning for a month.

This page runs one bottling line under two scheduling rules. The first is a fixed cycle that works like a train timetable: if a run is missed, the next slot for that product is already known. The second re-plans each week around whichever products are closest to running out. Both get the same orders, the same hours & the same disruption.

The line makes 100 SKU - stock-keeping unitOne specific sellable item. A 750 ml bottle & a 1 litre bottle of the same liquid count as two, because each is stocked, counted & ordered separately.SKUs with a typical Pareto profileA small share of items accounts for most of the volume, often summarised as 80/20. Because the items differ so much in size, they need to be planned differently.Pareto profile: 20 A, B & C itemsBands of the range sorted by volume. A items are the high-volume runners & C items the slow tail. A items are made often & C items rarely.A items carry 70% of volume, 30 B items carry 20% & 50 C items share the last 10%. Neither rule prevents the disruption. The comparison is about how quickly & how steadily each line recovers once it has passed.

A simulator by Alvin J Lin  ·  alvinlinjr@gmail.com  ·  open source under the MIT License

The timetableEach SKU has a fixed place on a repeating loop. A items appear on every turn, B items on every second turn & C items on every fourth. The ring turns at the run rate from the simulation below.
The chaseEach run goes to the SKU with the least cover at that moment, & each red line marks one changeover. Dots turn amber when an item is close to running out, & red when it has run out & customers are still waiting for orders.
Choose a disruption

01Before we start

How the model is set up

1

One line, 100 products

The line has 132 working hours a week & 100 SKUs to make. About 100 of those hours go on production to meet demand, which leaves roughly 32 for changeovers & for catching up after a problem.

2

Changeovers use line time

Switching from one product to the next means clearing down, cleaning & resetting the line. This Changeover, or setupClearing down, cleaning, changing parts & checking the first output of the new product. The line makes nothing sellable while it happens.changeover takes 45 minutes, & nothing sellable is made while it happens. Twenty changeovers use 15 hours a week.

3

A, B & C items

Twenty products carry 70% of sales, the next 30 carry 20% & the remaining 50 share 10%. High-volume items are made often & low-volume items rarely, so that each batch is large enough to justify its changeover.

Cumulative share of sales against share of products, for the 100 SKUs in the model. The dashed line shows what equal sales would look like.
4

Two scheduling rules

The wheel keeps a fixed sequence that is published weeks ahead. Reactive re-planning rebuilds each week's plan around the SKUs closest to running out. Reactive sounds more responsive, & the model tests whether that holds up.

5

Where the difference appears

Both rules lose output during the disruption & neither can avoid it. They separate in the weeks after it ends, when the missed volume has to be made in hours that current demand already needs.

To start, choose Supplier shortage & press Play weeks. Terms with a dotted underline open a short definition. The glossary & the maths are at the end of the page.

02The railway comparison

Why a timetable is most useful when a train is cancelled

On the top line every fifth train is cancelled, yet the sign always shows when the next one leaves. The bottom line never cancels a train. It sends one whenever the platform looks full, so trains bunch & nobody can say when the next will come. Step through the eleven points below & the animation will show each one.

Choose a step, or play the story
03Inside one week

The week, hour by hour

Each bar shows one week of line hours. Solid blocks are production, grey hatching is planned changeover time & red hatching is the extra time a rushed changeover takes. Press play to step through the weeks, or hover over a block to see what ran.

Week 0 Before
A run B run C run Planned changeover Extra time, rushed changeover Hours lost to disruption Spare
Scheduled serviceFixed product wheel WK 00
Unscheduled serviceReactive re-planning WK 00
04Result

The two plants compared

Both plants receive the same orders, hours & disruption. The only difference is how each decides what to run next.

05Service over time

Service during & after the disruption

Case fill rateThe share of cases ordered that shipped on time. Because it counts cases, the high-volume A items dominate it.Case fill each week, meaning the share of ordered cases shipped on time. The red band marks the disruption. For the reactive plant, most of the lost service comes after the band ends.

Product wheel Reactive re-planning Recovered: 97% or better for 4 weeks running Hover to read · click to go to that week

Changeover hours per week
The wheel's changeover hours follow the schedule, so they stay level & can be seen weeks ahead. Re-planning adds unscheduled changeovers, often before materials are staged, & these increase at the point where the line most needs its hours.
Orders carried forward
Volume that misses its week is carried into the next, where it competes with that week's demand for the same hours. This is why the reactive plant's worst week can fall after the disruption has ended.
06Stock by SKU

Where the stock sits in week 0

Each column is one SKU, with A items at the front & C items at the back. Column height shows weeks of CoverHow long current stock would last at normal sales. Two weeks of cover means the item runs out in two weeks if nothing more is made.cover. Red columns below the floor are SKUs that have run out while customers are still waiting for orders; the deeper the column, the more orders are waiting. Drag either model to rotate both.

Week 0 Before
Product wheel
Reactive re-planning
Out of stock, with customer orders still to ship Under 1 week of cover 1–2 weeks 2–4 weeks 4 weeks or more Overstocked: more than its cycle needs
07Line hours

How each plant spends its hours

Demand fixes the production hours. The rest go on changeovers or stay spare, & only the spare hours are available to absorb a disruption.

Average week, as run

Product wheel
Reactive re-planning

Why the last few hours matter most

Queueing delay rises with ρ / (1 − ρ), where ρ is the share of hours in use. Close to 100%, each extra hour of changeover adds far more delay than the hour before it. Appendix §7 gives the formula.

08The trade-off

What each rule gets for its stock & its hours

Every point on this chart is a full 78-week run with one setting changed, so each line shows the range of results a rule can reach. Lines higher on the chart deliver better service for the same stock or the same hours. Click any point to load its settings into the rest of the page.

09Assumptions

Change the assumptions

Each assumption in the model can be changed here. Available hours & the unplanned setup factor have the largest effect on the result, so set those two from your own changeover data before relying on it.

10Reading the result

What happened in this run

    Disruptions will reach the line under either rule. With a fixed cycle, the line meets them with more spare hours & a sequence it can return to.

    11Scope of the model

    What the model covers

    12Reference

    Glossary

    Every term used on this page.

    SKU stock-keeping unitOne specific sellable item. A 750 ml bottle & a 1 litre bottle of the same liquid are two SKUs, because each is stocked, counted & ordered separately.
    Pareto profileA pattern where a small share of items accounts for most of the volume, often summarised as 80/20.
    A, B & C items ABC classificationThe range sorted into bands by volume. A items are the high-volume runners & C items the low-volume tail.
    Changeover or setupClearing down, cleaning, changing parts & checking the first output of the next product. The line makes nothing sellable during it.
    Product wheelA fixed, repeating production sequence in which every SKU has a known next slot, much like a train timetable.
    Reactive re-planningRebuilding the schedule each week around whichever SKUs have the least stock left.
    EPEI every product every intervalThe time the line takes to make every SKU once & return to the start. A shorter EPEI means smaller batches & less stock, but more changeovers.
    Level loadingArranging the sequence so each week has a similar amount of work.
    CoverHow long current stock would last at normal sales. Two weeks of cover means the item runs out in two weeks if nothing more is made.
    Out of stock, orders waiting backorderA SKU has run out while customers still have orders for it. Those orders wait until the line makes more, & ship late.
    OverstockedHolding more of a SKU than its own cycle & safety stock call for.
    Cycle stockStock held because items are made in batches. A household that shops every two weeks keeps about two weeks of food.
    Safety stockExtra stock held for weeks when demand runs above average.
    Case fillThe share of ordered cases shipped on time. Because it counts cases, the high-volume A items dominate it.
    Complete order lines OTIF, on time in fullThe share of order lines delivered complete & on time. Every line counts the same, whatever its volume.
    Backlog orders carried forwardAll orders received but not yet shipped. They roll into later weeks & use hours those weeks need for their own demand.
    ExpediteMoving a job up the queue after an escalation. Work behind it is delayed, & the changeover often takes longer because it was not planned.
    Fire-fightingRunning the week from a priority list that keeps changing. Each decision may be sensible on its own, but together they use a lot of line time.
    Frozen horizonThe weeks of the schedule held fixed so that materials & labour can be arranged within normal lead times.
    RunnerA high-volume SKU with a permanent place in every turn of the wheel. These are the A items.
    UtilisationThe share of available line hours already in use. The higher it is, the fewer hours are left to recover from a problem.
    Schedule keptThe share of runs made as originally published.
    Coefficient of variationHow much demand moves around its average. A CV of 0.6 means weekly demand typically varies by about 60% of the average.
    Lead timeThe time from starting production to having stock ready to ship. The model uses one week throughout.
    ELSP economic lot scheduling problemThe standard problem of deciding how often to make each product on a shared line, balancing changeover cost against the cost of holding stock.
    13Appendix

    The maths behind the model

    Four relationships drive the model: the balance of hours on the line, the shortest cycle the line can sustain, how often each class of SKU should run, & what happens to stock when the gap between runs becomes unpredictable.

    Each section gives the idea in plain language, then the formula, then what it means on the line. The figures in the tables are calculated from the current settings, so the formulas can be checked against the simulation.

    Dtotal demand, units per week
    rrun rate, units per hour
    Havailable line hours per week
    splanned setup time, hours
    kunplanned setup multiplier
    NcSKUs in class c
    Tcreplenishment interval, weeks
    μimean weekly demand, SKU i
    σiits standard deviation
    Lproduction lead time, 1 week
    Λsetup load, hours per week
    ρutilisation of available hours
    Btbacklog at week t
    Qitrun quantity, SKU i week t

    §1 How demand is generated

    Weekly demand for each SKU is drawn from a normal distribution truncated at zero. The Coefficient of variationVariability measured against the average. A CV of 0.6 means weekly demand typically moves by about 60% of its average, which is common for slow movers.coefficient of variation rises as volume falls, & each C item has an 18% chance of no orders in a given week, which gives the tail its uneven pattern.

    dit = max(0, μi(1 + cvi·Zit)),   Zit ~ N(0,1) iid(1)

    Volume is split twice: across the bands by the 70/20/10 shares, then within each band by lognormal weights. Strictly this is Pareto-like rather than a true power law, because the band shares are set in advance. Items are then re-sorted into bands by their actual volume, as a real ABC review would do. Opening stock depends on where each item sits in the cycle, so the run starts in a settled state. Both rules start from the same position & see the same orders.

    What this means in practice

    Each SKU has a stable average, but smaller SKUs order more erratically. Most of the variability sits in the C items, which is also where the reactive rule does most of its damage.

    §2 How line capacity is used

    The hours used must fit within the hours available. This is the model's only hard constraint.

    ∑i∈St ( si + Qit/r )  ≤  Ht(2)

    On a wheel, the set of items run each week, St, comes from the schedule, so setup load is a constant set by the run frequencies:

    Λwheel = ∑c Nc·s / Tc(3)

    The week therefore divides into three parts:

    What this means in practice

    Only the production hours make anything. Setup hours are the cost of making many different products, & the slack is what remains to absorb a disruption. On a wheel the setup hours are fixed in advance, so the slack is known. Under re-planning they depend on events, & they tend to rise when the line is already under pressure.

    §3 The shortest cycle the line can sustain

    If all N items ran on one common interval T, each turn of the cycle would need N setups plus the production itself:

    N·s + (D/r)·T  ≤  H·T   ⇒   T  ≥  N·s / (H − D/r)  =  Tmin(4)

    Writing ρrun = D/(rH) for the share of hours spent producing:

    Tmin = N·s / ( H·(1 − ρrun) )(5)
    What this means in practice

    As the line gets busier, the shortest cycle it can keep gets longer, & it lengthens sharply as utilisation approaches 100%. On a busy line it is not possible to run every item more often, which is why §4 gives A, B & C items different intervals.

    §4 Why high-volume SKUs run more often

    Solving the ELSP - economic lot scheduling problemThe standard problem of deciding how often to make each product on a shared machine, balancing changeover cost against the cost of holding stock.economic lot scheduling problem item by item balances setup cost against holding cost, & gives an interval inversely proportional to the square root of demand:

    Ti* = √( 2s / (h·μi) )   ⇒   Ti ∝ μi−1/2   ⇒   Tc/TA = √(μA/μc)(6)
    What this means in practice

    Doubling the interval at each step from A to C is close to the lowest-cost pattern for a 70/20/10 split, so the wheel's frequencies can be justified on cost as well as on discipline. Running C items as often as A items would spend changeover hours that the high-volume items need.

    §5 The cycle stock created by making in batches

    Stock made to cover an interval is used up over that interval, so average Cycle stockStock held because items are made in batches rather than continuously. A household that shops every two weeks keeps about two weeks of food.cycle stock is half the run quantity:

    Icycle = ∑i μiTi/2 = (D/2)·∑c wcTc     cover = Icycle/D + SS(7)
    What this means in practice

    A household that shops every two weeks keeps about two weeks of food in the house; shopping weekly halves the stock but doubles the trips. The wheel holds more stock for the same reason, & that extra stock is the cost of running to a schedule.

    Cycle stock depends on how often we choose to run each item, while safety stock protects against variation in demand. Plants with high stock & poor service often have plenty of the first & too little of the second. Because a fixed cycle makes cycle stock predictable, the safety stock on top of it can be sized with some confidence.

    §6 Why a known next slot reduces safety stock

    With a fixed interval R & lead time L, safety stock covers the protection interval in the usual way:

    SS = z·σd·√(R + L)(8)

    If the interval itself varies, the variance gains a second term:

    SS = z·√( (R̄+L)·σd2  +  μd2·σR2 )(9)
    What this means in practice

    On a wheel, σR is close to zero because the next slot is published, so the second term drops out & safety stock only has to cover demand. Under re-planning, the timing of a SKU's next run depends on shortfalls elsewhere in the range. σR is then large & hard to estimate, which makes the buffer hard to size.

    The main benefit of a fixed cycle is that the gap between runs stays the same, even when that gap is fairly long. A rail timetable gives passengers the same assurance: trains still run late, but people know when the next one is due.

    §7 Why high utilisation amplifies disruption

    Utilisation includes setup time, so a rule that changes over more often runs a busier line on the same equipment:

    ρ = ( D/r + Λ ) / H(10)

    Kingman’s approximation then shows how waiting time responds, where ca & cs are the coefficients of variation of arrivals & service:

    W  ≈  ( (ca2 + cs2) / 2 ) · ( ρ / (1 − ρ) ) · τ(11)
    What this means in practice

    Reactive re-planning raises both terms. ρ rises because changeovers take more of the week, & cs2 rises because changeovers now vary between short planned ones & long rushed ones. With ρ/(1−ρ) in the formula, delay grows much faster than either input. Reducing available hours from 140 to 125 changes the inputs gradually but changes the result sharply.

    §8 How a backlog slows the line down

    Backlog grows by the difference between demand & output, & output is whatever the hours left after setup can produce:

    Bt+1 = Bt + Dt − Yt,     Yt = r·(Ht − Λt)(12)

    Under re-planning the number of expedited setups, et, rises with the backlog. Each one takes k times the planned setup, so output falls as the backlog grows:

    ∂Yt/∂Bt = −r·s·(k − 1)·∂et/∂Bt  <  0(13)

    On a wheel, setup load comes from the schedule & does not respond to the backlog:

    ∂Λt/∂Bt = 0   ⇒   ∂Yt/∂Bt = 0(14)
    What this means in practice

    Under re-planning, a growing backlog brings more rushed changeovers, which reduce output & let the backlog grow further. If that loop is strong enough the backlog keeps rising, which shows up as the sawtooth in the service chart. The bullwhip effect works through a similar loop, & in both cases the upset lasts far longer than the event that caused it.

    On a wheel, output in a bad week is the same as in a good one, so the backlog falls at a steady rate until it is cleared.

    §9 Case fill & line fill measure different things

    Case fill rateThe share of cases ordered that shipped on time. It is weighted by volume, so the A items dominate it.Case fill weights each SKU by its volume. Line fill, or OTIFThe share of order lines delivered complete & on time. Every SKU counts the same, so a customer who orders ten items & receives nine has a failed line even though 90% of the units arrived.Line fill weights every SKU equally, so the two measures can give very different answers:

    CFRt = ∑i min(xit, dit) / ∑i dit      LFRt = |{ i : xit ≥ dit }| / |{ i : dit > 0 }|(15)
    What this means in practice

    C items make up half of the order lines but only a tenth of the cases. A plant can report good case fill while a customer ordering across the range finds half their lines short. When reactive re-planning drops C items, case fill hardly moves but OTIF falls, so judging the two rules on case fill alone understates the gap between them.

    §10 What the model leaves out

    Some of these omissions favour the wheel & some count against it.

    • There is one line, with no parallel resources to share the load.
    • Setup time does not depend on sequence. Real wheels are sequenced to shorten changeovers by grade, colour or allergen, so a real wheel would gain more than this one. Understates the wheel.
    • Lead time is a flat one week, with no supply variation apart from the disruption itself.
    • All unmet demand is backordered, with no lost sales or substitution. Lost sales would hurt both rules, & reactive more.
    • Demand is independent across SKUs & weeks, with no seasonality, promotions or forecast bias.
    • Capacity is fixed, with no overtime, extra shifts or subcontracting. A plant that is fire-fighting usually buys extra hours, which moves the cost onto the P&L.
    • Inventory records are accurate, & there is no yield loss apart from the quality hold.
    • The wheel is assumed to be followed. Holding to it under commercial pressure is the hardest part in practice, & the model does not test it. Favours the wheel.

    The minimum economic run length & the unplanned setup factor k carry most of the weight. Both can be measured in a real plant, so set them from your own changeover studies before using the model to support a decision.

    Created by

    Alvin J Lin

    alvinlinjr@gmail.com

    Licence

    Open source under the MIT License. You are free to use, copy, modify, merge, publish, distribute, sublicense & sell this work, including commercially, provided the copyright notice & the licence text travel with it. It is provided without warranty of any kind.

    Read the full licence
    MIT License
    
    Copyright (c) 2026 Alvin J Lin
    
    Permission is hereby granted, free of charge, to any person obtaining a copy
    of this software & associated documentation files (the "Software"), to deal
    in the Software without restriction, including without limitation the rights
    to use, copy, modify, merge, publish, distribute, sublicense, &/or sell
    copies of the Software, & to permit persons to whom the Software is
    furnished to do so, subject to the following conditions:
    
    The above copyright notice & this permission notice shall be included in
    all copies or substantial portions of the Software.
    
    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
    FITNESS FOR A PARTICULAR PURPOSE & NONINFRINGEMENT. IN NO EVENT SHALL THE
    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
    FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
    DEALINGS IN THE SOFTWARE.

    Reuse & adaptation

    The whole simulator is one self-contained HTML file with no dependencies: every animation & 3D view is drawn by hand on canvas. Every model assumption sits in a single configuration block near the top of the script, & the appendix recomputes itself from whatever you change. If you adapt it with your own plant data, the attribution above is all that is asked.

    Cycle or Chase · MIT License © 2026 Alvin J Lin · alvinlinjr@gmail.com