The gold standard of yard management is a service plan that remains understandable when arrivals bunch together or one visit takes longer than expected. Average daily capacity is only a starting point. A receiving process can have enough time over the whole day and still develop a severe queue during one concentrated hour.
This guide explains arrival clustering and queue recovery through original fictional schedules. You will trace waiting time visit by visit, distinguish spare time from usable recovery capacity, test a temporary outage and check whether an apparent improvement merely moves delay to another boundary. The calculations concern an abstract receiving service. They do not direct vehicle movements or establish an operating layout.
The Yard Management subject guide provides the wider identities, handovers and responsibilities. The existing Delivery Appointment Management guide owns the appointment as a coordinated promise. This article develops a narrower quantitative question: why queues form despite favourable averages, how delays propagate and what evidence can demonstrate recovery.
eduKateSG · Yard Management
Find your next learning step
Choose a task-based route, then follow the expanded teaching chapters below.
Chapters 1–3 · Define the model
Separate visits, workload and service starts.
Chapters 4–7 · Explain delay propagation
Compare clusters, queue burden and recovery gaps.
Chapters 8–12 · Test constraints and consequences
Examine outages, compatibility and service rules.
Chapters 13–15 · Verify and transfer
Reconcile records, practise and preserve opening work.
Chapters 16–17 · Questions and specialist routes
Chapter contents
Define the model · Chapters 1–3
Explain delay propagation · Chapters 4–7
Test constraints and consequences · Chapters 8–12
Verify and transfer · Chapters 13–15
Questions and specialist routes · Chapters 16–17
A queue is a set of work waiting for a specified service under a stated rule. The definition needs a boundary. Vehicles outside a site, visits checked in but not served, trailers awaiting a particular interface and consignments awaiting a decision are different populations. They should not be merged into one number without an explanation.
For the fictional Cedar Depot model, each visit requires one abstract receiving service. Arrival means joining the modelled waiting population. Service start means beginning that service. Completion means ending it. The model includes no gate processing, physical manoeuvre, unloading detail or later departure stage.
Those omissions are deliberate boundaries for the arithmetic. An actual operation needs the relevant stages and competent arrangements. A model can teach the timing relationship without pretending that one service interval describes the complete yard.
Now define the service rule. The first examples use one server, one visit at a time and first-come, first-served order. All visits are eligible for that server. These assumptions make the sequence reproducible. A real priority or compatibility rule would produce a different model and needs to be stated.
Record a visit identity with each time. A corrected arrival entry is not another visit. A vehicle returning later may be another visit under the chosen definition. The same identity discipline that supports the root article also protects queue calculations from duplicated work.
The first useful question is therefore small: how many defined visits are waiting for this particular service at this particular time? Once that is clear, waiting duration, workload and recovery can be calculated on compatible terms.
CHAPTER 2 OF 17 · Define the model
2. Separate arrivals, workload and available service
An arrival count says how many visits enter the model. Workload says how much service those visits require. Available service says what the receiving process can provide over a defined period under the model’s assumptions. Equal arrival counts do not establish equal workload.
Suppose six visits each require twenty minutes of the same service. Their total workload is 120 server-minutes. One continuously available server can provide 120 server-minutes over two hours. The arithmetic appears balanced, but it says nothing yet about when the visits arrive.
If all six arrive at the beginning, the last waits one hundred minutes before service. If they arrive twenty minutes apart, each can start immediately under this deterministic model. The same work and the same total availability produce different waiting experiences.
Capacity also needs a relevant resource. An additional waiting position is not another receiving server. An extra server that cannot perform the required service is not interchangeable capacity. A two-hour total can conceal a one-hour outage inside the period.
Write separate lines for demand and availability. Demand includes arrival times, required service and eligibility. Availability includes the server or resource, usable periods and relevant restrictions. The relationship between those lines produces the queue; neither line alone describes it.
For students, use cards representing visits and paper strips representing service minutes. This is a safe abstract scheduling exercise. It makes the difference between counting jobs and counting work visible before introducing uncertainty or software. The useful question is how the work fits through time, not simply whether the daily totals look comfortable.
For one eligible server, a visit can start when it has arrived and the previous visit has completed. Its start is the later of those two times. Waiting time is start minus arrival. Completion is start plus the supplied service duration. This recurrence is ordinary timing arithmetic under the stated model.
Consider visits A, B and C arriving at minutes 0, 10 and 20. Each requires twenty minutes. A starts at 0 and completes at 20. B has arrived by then, so it starts at 20, waits ten minutes and completes at 40. C starts at 40, waits twenty minutes and completes at 60.
| Visit | Arrival minute | Service duration | Start minute | Completion minute | Wait |
|---|---|---|---|---|---|
| A | 0 | 20 min | 0 | 20 | 0 min |
| B | 10 | 20 min | 20 | 40 | 10 min |
| C | 20 | 20 min | 40 | 60 | 20 min |
The total wait is thirty visit-minutes and the mean wait is ten minutes. The last visit completes forty minutes after its arrival. That elapsed time includes twenty minutes waiting and twenty minutes service.
Ask the learner to reconstruct B without looking at the answer. The server’s previous completion, rather than B’s appointment label, determines its earliest modelled start. Then change C’s arrival to minute 50. It starts at 50 after an idle interval, waits zero and completes at 70.
This small change checks whether the learner understands the relationship. A memorised pattern of adding ten minutes of waiting to every next visit fails as soon as the arrival sequence changes.
CHAPTER 4 OF 17 · Explain delay propagation
4. Show why an average can hide a cluster
Now compare two six-visit schedules over a two-hour period. Every visit requires twenty minutes. In the evenly spaced schedule, arrivals occur at minutes 0, 20, 40, 60, 80 and 100. Each starts immediately and the last completes at 120.
In the clustered schedule, three visits arrive at minute 0 and three at minute 60. Within each simultaneous group, a supplied A-before-B-before-C rule determines order. Waiting times in each cluster are 0, 20 and 40 minutes. Across all six visits, the mean wait is twenty minutes and total wait is 120 visit-minutes.
Both schedules contain six visits and 120 minutes of work. Both complete by minute 120 under the model. One produces no waiting; the other produces a maximum wait of forty minutes. The daily or period throughput does not reveal that difference.
| Model schedule | Visits completed | Total service work | Mean wait | Maximum wait |
|---|---|---|---|---|
| Even spacing | 6 | 120 server-minutes | 0 min | 0 min |
| Two clusters | 6 | 120 server-minutes | 20 min | 40 min |
The lesson is not that every real site should use twenty-minute appointments. The values are invented to isolate clustering. Real service durations vary, resources differ and arrival arrangements must fit the actual operation.
A useful report therefore shows arrival pattern and service pattern alongside totals. If a manager sees enough daily capacity but drivers report waiting, both accounts can be true. The missing information may be when demand reached the constraint, not whether anyone calculated the daily average correctly.
CHAPTER 5 OF 17 · Explain delay propagation
5. Measure the waiting burden through time
Total waiting can be calculated by adding each visit’s wait. It can also be understood as the area beneath a queue-length graph. If two visits wait for twenty minutes, that interval contributes forty visit-minutes. The unit combines a count with a duration.
For the first cluster of three simultaneous arrivals, one visit begins immediately. Two wait during minutes 0–20, contributing forty visit-minutes. One waits during minutes 20–40, contributing twenty more. Total waiting for the cluster is sixty visit-minutes, matching 0 + 20 + 40.
The second cluster adds another sixty. Over the complete 120-minute observation window, the mean waiting population is 120 visit-minutes divided by 120 minutes, or one visit. This is a time average of the queue, not the average number arriving in a batch.
Keep service and waiting populations separate. A graph counting visits in service as well as waiting describes work in the system. Its area relates to total time in the system under consistent boundaries, rather than waiting alone. A dashboard can appear inconsistent simply because two teams use different definitions of present.
The identity between summed waits and queue area provides a useful reconciliation check when all relevant visits and intervals are included. If the two calculations disagree, inspect boundary conventions, missing events, simultaneous timestamps and unfinished work.
This is a strong learner exercise because the same result appears in two representations: a visit table and a time graph. Understanding their agreement makes the method more robust than treating the spreadsheet total as an unexplained final answer.
CHAPTER 6 OF 17 · Explain delay propagation
6. Follow one overrun into later visits
A long service can create waiting for later visits even when those visits arrive exactly as planned. Trace the sequence before assigning a cause to the people who are waiting.
Six visits arrive at minutes 0, 20, 40, 60, 80 and 100. Their service durations are 10, 10, 50, 10, 10 and 10 minutes. A and B complete at 10 and 30, leaving idle gaps before the next arrivals. C starts at 40 and completes at 90. D therefore waits thirty minutes, E waits twenty and F waits ten. F completes at 120.
| Visit | Arrival | Duration | Start | Completion | Wait |
|---|---|---|---|---|---|
| A | 0 | 10 | 0 | 10 | 0 |
| B | 20 | 10 | 20 | 30 | 0 |
| C | 40 | 50 | 40 | 90 | 0 |
| D | 60 | 10 | 90 | 100 | 30 |
| E | 80 | 10 | 100 | 110 | 20 |
| F | 100 | 10 | 110 | 120 | 10 |
The modelled mean wait is ten minutes. The total workload is one hundred server-minutes over a 120-minute completion period. Twenty minutes were idle earlier, but that earlier spare time could not be stored and used simultaneously with C’s later service.
This explains why spare time must be located, not merely totalled. The recovery after C occurs because later ten-minute services are shorter than the twenty-minute arrival gaps. Each later completion gains ground until the backlog clears.
The result does not establish why C required fifty minutes. That explanation needs its actual service and exception record. The timing model shows propagation; it does not diagnose the original operational cause.
CHAPTER 7 OF 17 · Explain delay propagation
7. Place a recovery gap where it can do useful work
A buffer is useful when it protects a defined relationship against a relevant variation. Spare minutes placed far from the disruption may provide little recovery. The schedule needs to show where and why the margin exists.
Keep the same six service durations, but move D, E and F to minutes 100, 120 and 140. C still completes at 90. The later visits can now start on arrival and none waits under the supplied model. The last completes at 150, thirty minutes later than in the previous schedule.
The change has traded a later planned sequence for less modelled on-site waiting. It has not created more physical capacity. Whether that trade is useful depends on the complete delivery obligations, carrier arrangements and other resources. A planner should not announce an improvement from the queue figure alone.
Compare at least three outcomes: waiting experienced after arrival, the time by which all work completes and the effect on the promised service. If moving an appointment simply forces a vehicle to wait elsewhere after it has already committed to the journey, the total burden may remain.
Distinguish planned margin from a guaranteed duration. A ten-minute buffer can absorb some variation, but it does not mean every overrun will be ten minutes or less. Select and review buffers through relevant evidence and consequences rather than treating one percentage as universal.
For a paper exercise, give students a limited amount of spare time to distribute across an invented schedule. Then introduce a disruption at different points. Ask which placement helps and which does not. They should explain the relationship between timing, the constraint and the next commitment, rather than choose the arrangement with the most empty calendar space.
CHAPTER 8 OF 17 · Test constraints and consequences
8. Test a temporary loss of capacity
Availability can change within a day. A server or receiving interface unavailable for a defined period creates a different schedule from one that is continuously available, even if an average capacity figure remains unchanged on the dashboard.
Consider three visits arriving at minutes 0, 20 and 40, each requiring twenty minutes. Under continuous availability, none waits. Now suppose the abstract service is unavailable from minute 20 to minute 50. A completes at 20. B cannot start until 50 and completes at 70. C starts at 70 and completes at 90. Their waits are 0, 30 and 30 minutes.
The outage shifts later work, and the queue persists after availability returns. Reopening the server at minute 50 does not instantly remove waiting visits. Capacity restoration and queue recovery are separate events.
An actual restriction belongs to the competent site’s process. The scheduling record needs the resulting availability and its effective time. A booking calendar should not continue promising unavailable service because a technical decision has not reached the planner.
For a changed-context check, move the outage to minutes 100–130, after these three visits have finished. The same thirty-minute loss has no effect on their waits. The amount of lost time is identical, while its location relative to demand changes the outcome.
That comparison helps learners see why a daily availability percentage can be inadequate for a time-sensitive service. The record needs the periods that matter to the actual arrivals and the service sequence, not only a total number of minutes marked available.
CHAPTER 9 OF 17 · Test constraints and consequences
9. Keep compatibility separate from the number of bays
Two servers do not automatically double usable capacity for every visit. A visit may require a particular interface, team, equipment arrangement or condition. The model must identify eligible resources rather than treat every empty slot as interchangeable.
In a fictional two-server exercise, Server 1 can handle categories R and S. Server 2 can handle only S. Two R visits and one S visit arrive together, each requiring twenty minutes. One R can use Server 1 while S uses Server 2. The second R waits for Server 1 even if Server 2 becomes idle later.
This is a compatibility constraint. It does not disappear because the total number of servers is two. A plan that assigns the second R to Server 2 creates apparent capacity by violating its own assumptions.
The next question is whether the resource allocation is sensible within the stated model. If an S visit occupies Server 1 while Server 2 is available and eligible for S, it may delay R work unnecessarily. But a real reassignment needs the appropriate operational assessment and authority; a paper scheduling argument cannot establish physical suitability.
Record eligibility explicitly. Use a matrix with visits or categories on one side and resources on the other. Mark what is confirmed, what is unavailable and what remains uncertain. Do not convert uncertain eligibility into a yes simply to make the schedule balance.
The Warehouse Dock Leveller Maintenance guide provides one specialist route into why an interface needs its own condition evidence. Queue arithmetic should use confirmed availability while preserving the specialist responsibility behind it.
CHAPTER 10 OF 17 · Test constraints and consequences
10. Treat uncertain arrivals as scenarios, not promises
An estimated arrival time is a prediction. An appointment is an agreed service arrangement. An actual arrival is an observed event under a definition. Keeping them separate makes uncertainty easier to manage and later performance easier to explain.
A useful planning exercise tests several plausible arrival patterns rather than relying on one perfectly spaced sequence. The baseline may use expected times. Another scenario can cluster several arrivals. A third can combine a late arrival with a longer service. These are conditional outcomes under supplied assumptions, not predictions of exact future waits.
Suppose a plan works when visits arrive twenty minutes apart but develops a long queue when three arrive together. The useful conclusion is that the plan is sensitive to clustering. The next investigation asks how frequently that pattern occurs, whether information can arrive earlier and which response is actually available.
Do not attach invented probabilities to scenarios. If a classroom exercise has no evidence about frequency, label the cases as possibilities. A worst-looking case is not automatically a realistic worst case; it is simply one defined scenario until the relevant boundaries and evidence support a stronger claim.
Preserve update lead time. A revised estimate received before a carrier commits to the site can support a different response from the same information received at the gate. The existing Delivery Appointment Management article develops that coordination role.
For students, compare the same schedule with information arriving thirty minutes before a disruption and thirty minutes after it. Ask what decisions remain possible in each case. Timing information is useful because it changes the available action, not because a dashboard contains another timestamp.
CHAPTER 11 OF 17 · Test constraints and consequences
11. Avoid improving the number by moving the boundary
A shorter reported wait can come from a better process or a changed definition. If the clock starts later, some waiting may disappear from the metric while remaining in the driver’s day. A fair comparison keeps the measurement boundary visible.
Imagine the first report measures arrival at the designated queue to service start. The next measures completed check-in to service start. If check-in itself takes twenty minutes, the second figure can be twenty minutes shorter without any earlier service. Both intervals can be useful, but they should not be presented as the same outcome.
Likewise, holding arrivals outside the observed area may reduce the visible queue. The complete service question asks where the time went, which party bears it and whether the change was communicated through an appropriate arrangement. A smaller on-site count is not automatically a smaller total burden.
Keep unfinished visits in view. If the reporting period ends while the longest waits remain unresolved, averaging only completed visits can make a difficult day appear favourable. Report the open population and its elapsed waits alongside completed-visit statistics, with the method clearly described.
Separate targets from evidence. A planned maximum wait is a promise or objective; an observed maximum comes from the records included. The fact that a plan contains a target does not establish that every visit met it.
For an evidence check, ask learners to compare two reports with identical service events but different start timestamps. They should identify the definition change before discussing improvement. That habit transfers to healthcare waiting lists, customer support and school administration, where a convenient clock can otherwise conceal the experience the user actually had.
CHAPTER 12 OF 17 · Test constraints and consequences
12. Choose a recovery rule with consequences in view
When a queue exists, the order of service becomes a consequential choice. First-come, first-served is easy to explain, but actual operations can have other justified priorities and constraints. The rule needs appropriate governance rather than an improvised decision made whenever a different visit appears urgent.
In a paper model, compare two eligible visits waiting at the same time. One requires ten minutes, the other forty. Serving the shorter first reduces their combined waiting under some conditions, but it may conflict with a service commitment or a consequence attached to the longer visit. A numerical objective is only part of the decision.
State the objective precisely. Minimising mean wait, protecting a particular deadline, reducing the longest wait and preserving a special condition are different goals. A rule that improves one can worsen another. The plan should identify which outcomes matter and which conditions cannot be traded away through a convenient score.
Actual load, equipment, safety and custody decisions remain with the responsible competent process. HSE’s Delivering Safely guidance emphasises cooperation among supplier, carrier and recipient in its UK context. It supports the limited principle that the interface needs agreed responsibility; it does not prescribe Cedar Depot’s fictional queue rule.
For learning, ask students to compare the outcomes under two explicitly supplied rules. They should explain whose wait changes and what information would be needed to judge the choice. The goal is transparent reasoning about consequences, not pretending that one scheduling slogan resolves every operational obligation.
CHAPTER 13 OF 17 · Verify and transfer
13. Verify the model against the visit records
A model should be reproducible from its inputs and checked against the relevant event records. Start with identities, arrivals, service starts, completions, availability and eligibility. Preserve corrections rather than replacing earlier values without explanation.
For each visit, verify that start is no earlier than arrival, completion is no earlier than start and the recorded service duration matches the chosen interval. For a single-server model, confirm that eligible service intervals do not overlap. These checks test the model’s own consistency, not the safety or accuracy of every real-world event behind the data.
Reconcile the population. Opening unfinished work plus arrivals minus completions should equal closing unfinished work under one definition and boundary. An internal location update does not add another arrival. A cancelled booking is not automatically a physical departure.
Then examine discrepancies between planned and observed times. A mismatch may arise from duration assumptions, unavailable resources, different event definitions or a missing visit. Select the next check from the pattern rather than adjusting whichever number makes the prediction look better.
For a pilot, define the comparison period and relevant outcomes before changing the schedule. Keep volume, work mix, availability and unresolved visits visible. A favourable before-and-after average does not isolate the schedule’s effect if the second period contained much easier work.
The root article’s discipline remains useful: distinguish intention, acknowledgement and confirmed event. A beautifully balanced schedule is a plan. A verified service account explains what happened, what remains open and why the next planning assumption deserves to change.
CHAPTER 14 OF 17 · Verify and transfer
14. Practise recovery thinking in another system
Begin the repair pathway with three visit cards and one service strip. Ask the learner to find starts and waits. If they add service duration to arrival without checking the previous completion, return to the single constraint: one server cannot serve two eligible visits at once under this model.
For stabilisation, mix evenly spaced arrivals, clusters, idle gaps and one longer service. The learner should choose the recurrence rather than memorise a table. Ask them to explain why earlier idle time cannot always absorb a later overrun.
For extension, introduce two compatible resources and one eligibility restriction. Require both a schedule and a statement of the assumptions. Then remove a resource for part of the period and ask which visits need revision. The learner should preserve the restriction rather than invent capacity to complete the table.
Change the setting to library equipment loans. The objects and responsibilities differ, but arrivals, service requirements, eligibility and recovery still matter. A camera being returned late can affect the next booking even if the week’s total demand is below nominal availability.
Finally ask the learner to critique a claim: the average daily workload is below capacity, therefore nobody should wait. A strong response constructs a small cluster counterexample and explains the time relationship. It does more than disagree with the conclusion.
This is the connection to The Gold Standard of Education. A learner gains a method for recognising a bottleneck, checking a quantity and explaining the limits of a reassuring average in a new practical setting.
CHAPTER 15 OF 17 · Verify and transfer
15. Include the opening backlog and the unfinished tail
A schedule does not necessarily begin with an empty system. If work is already waiting or in service when the observation period starts, a model that ignores it will predict starts too early. The opening state belongs among the inputs, alongside new arrivals.
Suppose the modelled server has twenty minutes of an existing service remaining at minute 0. New visits A, B and C arrive at minutes 0, 20 and 40, each requiring twenty minutes. A starts at 20 and completes at 40. B starts at 40 and completes at 60. C starts at 60 and completes at 80. Each new visit waits twenty minutes.
The same arrival and duration table would have produced zero waits in an empty system. The difference comes entirely from the opening work. A comparison between days therefore needs to preserve the initial state rather than assign every later delay to the day’s booking decisions.
Now stop reporting at minute 60. A and B are complete, while C begins service at the boundary and remains unfinished. If a report includes only completed new visits, it omits C’s later completion. In a more varied dataset, excluding unfinished visits can preferentially remove the longest cases.
Report both completed work and open work under a clear convention. A visit still waiting has an elapsed wait that may increase. A visit in service has a different status. Neither should be assigned a final duration before the event has occurred, and neither should disappear simply because the reporting window closed.
For the simple example, the new-arrival count is three and completed-new-visit count by minute 60 is two. One new visit remains in service immediately after that boundary under the chosen event convention. The original opening visit is a separate identity. Combining it with the new arrivals without explaining the population would distort the account.
Boundary conventions matter when several events share the same timestamp. State whether a completion at minute 60 belongs to the period ending at 60 and how the next period begins. The choice should avoid both omission and double counting. The physical event has not changed, but an inconsistent reporting rule can change apparent throughput.
For a learner exercise, provide two consecutive reporting windows and ask students to carry the closing unfinished population into the next opening state. Then introduce one repeated update to an existing visit. They should preserve the identity and state rather than count a fresh arrival.
This check connects the queue model with ordinary accounting for flows. Opening work, new arrivals, completions and closing work must describe the same population and boundary. Recovery is complete only when the relevant backlog has actually cleared under that definition, not when the reporting date changes or the unfinished rows move to a different tab.
CHAPTER 16 OF 17 · Questions and specialist routes
16. Questions about queues, buffers and improvement
Can there be a queue when daily capacity is sufficient?
Yes. Demand can reach a constraint faster than it can be served during part of the day. The clustered schedule shows this even with equal total workload and period availability.
Does another waiting position increase service capacity?
It changes where work can wait under the actual approved arrangements. It does not automatically increase the rate at which the required service can be completed. Waiting space and receiving service are different resources.
Does reopening a resource clear the queue immediately?
No. Waiting work still requires service. The temporary-outage example separates restored availability from later recovery of the backlog.
Is a lower average wait enough to prove improvement?
Check definitions, work mix, the delayed tail, unfinished visits and time moved outside the measurement boundary. The average is one useful summary rather than a complete account.
Should every appointment include the same buffer?
The useful margin depends on service variation, arrival patterns, consequences and available responses. The examples explain placement and propagation, not a universal buffer duration or percentage.
What is the smallest useful review?
Trace a small set of defined visits from arrival through service, using consistent timestamps and known availability. Reconstruct the waits, identify one propagation pattern and choose the next evidence check. Keep real operational decisions within the site’s responsible process.
CHAPTER 17 OF 17 · Questions and specialist routes
17. Continue into the specialist guides and source context
For the complete interface, read Yard Management, Warehousing and Logistics. For the coordinated service promise, use Delivery Appointment Management. For carrier-side resources, follow Fleet Management.
For the receiving interface and planning, read Staging and Dock Scheduling. For predictions and consequences, use ETA Prediction and Waiting Time in Logistics. Those guides own the broader receiving, forecasting and resource-cost questions; this article isolates queue propagation through reproducible schedules.
For interface and load conditions, use Warehouse Dock Leveller Maintenance, Refrigerated Transport and Cold Chain Companies. These routes own specialist conditions that a generic timing model cannot establish.
For a different scale of coordinated arrival, The Arrival-Time Agreement examines the maritime relationship. It is a separate port topic, not evidence that a warehouse service rule transfers unchanged to ships.
HSE’s linked delivery guidance and Visiting Drivers resource provide UK workplace-transport context. Singapore MOM's Risk Management page provides the official local starting route for workplace risk management. These sources were checked on 21 October 2026; no legal threshold or operational procedure is inferred from them here.
Every schedule and calculation in this article is an original fictional deterministic model. For wider learning routes, return to The Gold Standard of Education and the How X Works library.