Date: 5 October 2026 Publication Type: Research Project: MaMeeFarm™
Abstract
Production records usually preserve quantities, times, activities, interruptions, inputs, outputs, and other operational events.
Individually, these records describe what happened.
Connected through time, they can reveal something more: how work moved through the production system.
This distinction matters because production is rarely determined only by which activities are performed. The order in which those activities occur can influence waiting time, setup requirements, work in progress, resource conflicts, bottlenecks, rework, cycle time, quality, and final throughput.
Two production periods can therefore use substantially the same equipment, labor, activities, and inputs while producing different operating outcomes because work moved through the system differently.
The reverse is also possible.
Final output may remain stable even though the underlying sequence has become less efficient. Additional waiting, overtime, work in progress, buffering, or operator intervention may compensate for the changed sequence.
This research examines how longitudinal production records can make these changes visible. It focuses on activity order, timestamps, transitions, waiting, setup, interruptions, rework, bottlenecks, throughput, and output, while distinguishing observed sequence changes from causal claims about their effects.
The central finding is that production records reveal sequence effects most effectively when they preserve not only what work was completed, but the path through which work reached completion.
1. Research Question
How can longitudinal production records reveal the effects of changing work sequences when the same activities, resources, and production system remain available, and what evidence is required to distinguish sequence effects from changes in workload, timing, inputs, equipment condition, and operating environment?
2. Production Has an Order
A production record might show that five activities occurred:
A → B → C → D → E
Another production period may contain exactly the same five activities:
A → C → B → D → E
A simple activity count would describe these periods as equivalent.
They are not necessarily operationally equivalent.
Changing the order can alter:
- setup requirements;
- resource availability;
- waiting;
- handoffs;
- material condition;
- work in progress;
- rework;
- cycle time; and
- interactions with other jobs.
Production records therefore need to preserve sequence when sequence can affect performance.
3. Activity Presence and Activity Order Are Different Evidence
Suppose two production records show:
- 10 preparation activities;
- 10 processing activities;
- 10 inspections; and
- 10 completed outputs.
The counts are identical.
But the first process may perform inspection immediately after each processing activity.
The second may complete all processing before beginning inspection.
The activity totals cannot reveal this difference.
This illustrates a fundamental distinction:
activity occurrence
is not the same evidence as
activity sequence.
When sequence matters, aggregate counts can destroy information about how production actually occurred.
4. Timestamps Transform Activity Records Into Process Records
A list of completed activities describes what happened.
Timestamps can begin to reveal how work moved.
Useful temporal evidence can include:
- activity start;
- activity completion;
- handoff;
- queue entry;
- queue exit;
- machine start;
- machine stop;
- inspection;
- rework;
- failure; and
- final completion.
When these events can be connected to the same production unit, batch, job, asset, or process instance, researchers can reconstruct an observed sequence.
The production record becomes more than an inventory of actions.
It becomes a partial history of process flow.
5. Sequence Requires Identity
Time alone is insufficient.
Suppose records show:
09:00: Activity A 09:05: Activity B 09:10: Activity C
Researchers cannot automatically conclude that one production unit moved through A, B, and C.
The events may belong to different units.
Sequence reconstruction therefore requires some form of stable process identity.
Depending on the production system, this may be:
- batch;
- job;
- order;
- product;
- asset;
- work item;
- field operation; or
- another traceable unit.
Without identity, chronological proximity can be mistaken for process sequence.
6. Sequence Change Can Be Observed Directly
When sufficiently detailed records exist, researchers may observe that a recurring process changed from:
A → B → C → D
to:
A → C → B → D
This establishes a sequence change.
It does not yet establish a performance effect.
The next question is whether other production measures also changed.
For example:
- Did total completion time change?
- Did waiting increase?
- Did setup increase?
- Did rework change?
- Did throughput change?
- Did quality change?
- Did work in progress accumulate?
Sequence evidence and performance evidence should initially remain separate.
7. Waiting Time Can Reveal the Consequence of Order
An activity can take exactly the same processing time while total production time increases.
Suppose:
Activity A: 10 minutes Activity B: 10 minutes Activity C: 10 minutes
The direct work content is 30 minutes.
But if the production unit waits 20 minutes between A and B and 40 minutes between B and C, elapsed completion time becomes 90 minutes.
The additional time is not visible from activity duration alone.
Sequence records can expose:
processing → waiting → processing → waiting → processing
rather than merely:
three activities completed.
Waiting is therefore an important part of production history.
8. Lead Time and Processing Time Should Be Distinguished
Production records often become more informative when they distinguish:
time being worked on
from
time waiting to be worked on.
A process can have stable direct processing time while total lead time increases.
This can occur because of:
- queues;
- resource conflicts;
- batching;
- delayed handoffs;
- inspection delays;
- material availability; or
- changed scheduling.
If only processing time is recorded, a sequence related deterioration can remain invisible.
9. Sequence Can Change Setup Requirements
Production scheduling research has long recognized sequence dependent setup times.
The preparation required for the next task can depend on which task came before it.
Changing from Product A to Product B may require one setup.
Changing from Product C to Product B may require another.
The production consequence is therefore not determined only by the next activity.
It can depend on the transition between the previous state and the next state.
This means that production records capable of preserving transitions can reveal information that isolated activity records cannot.
10. Transition Records Can Reveal Hidden Costs
Suppose a machine produces three product types:
A, B, and C.
One sequence is:
A → A → B → B → C → C
Another is:
A → C → B → A → C → B
Both produce two units of each type.
Output composition is identical.
But the second sequence may require more changeovers.
If setup is sequence dependent, it may require more:
- cleaning;
- tool adjustment;
- material replacement;
- calibration; or
- preparation.
Final output counts alone cannot reveal this difference.
Transition history can.
11. Work in Progress Preserves Another Signal
Changing work sequence can alter how unfinished production accumulates.
If upstream activities proceed faster than downstream activities can accept work, work in progress can increase.
The final daily output may initially remain unchanged.
But the system begins storing unfinished work internally.
This can create:
- longer lead times;
- additional handling;
- space requirements;
- tracking complexity;
- quality exposure; and
- future congestion.
Work in progress is therefore not merely inventory.
It can be evidence of how production flow is behaving.
12. Bottlenecks Become Visible Through Repeated Flow Records
A bottleneck is not identified simply because an activity appears slow once.
Repeated records can reveal that work consistently accumulates before the same resource or activity.
Researchers may observe:
- queue growth;
- high utilization;
- long waiting times;
- delayed completion; or
- downstream starvation.
The bottleneck can also move.
A sequence change that improves one stage may shift congestion elsewhere.
This is why production performance should be studied as a system rather than as isolated activity efficiency.
13. Local Improvement Can Produce System Level Deterioration
Suppose one activity is reorganized to complete work faster.
Its local performance improves.
But it now sends work downstream in larger bursts.
The next activity cannot absorb the new arrival pattern.
Queues increase.
Total lead time grows.
The first activity looks better.
The production system performs worse.
Production records that preserve only local activity duration can miss this outcome.
Records that preserve flow across activities can reveal it.
14. Throughput Can Remain Stable While Flow Deteriorates
Consider two periods.
Period A
100 units completed Low work in progress Short queues Little overtime
Period B
100 units completed High work in progress Long queues Frequent overtime
Final throughput is identical.
Production behavior is not.
This leads to an important principle:
stable output does not establish stable process performance.
Sequence changes may first appear in the resources and time required to maintain output rather than in final output itself.
15. Throughput Can Change Without the Sequence Being Responsible
The reverse problem is equally important.
Suppose sequence changes and throughput subsequently falls.
This does not establish:
sequence change caused lower throughput.
During the same period:
- input quality may have changed;
- equipment may have deteriorated;
- environmental conditions may have changed;
- workload may have increased;
- staffing may have changed;
- maintenance may have occurred; or
- measurement methods may have changed.
The observed evidence is initially:
sequence changed + throughput changed.
Causal attribution requires more.
16. Production Records Can Reveal Sequence Variants
Real production rarely follows one perfectly fixed path.
Repeated records may reveal several variants:
A → B → C → D
A → C → B → D
A → B → D
A → B → C → B → D
These paths can represent:
- alternative procedures;
- exceptions;
- skipped steps;
- rework;
- interruptions; or
- changes in operating practice.
Their frequency can then be examined through time.
A previously rare sequence may become common.
A standard path may disappear.
The change in sequence distribution can itself become a research finding.
17. Rework Creates Loops
Production is not always a straight line.
A product may move:
processing → inspection → rework → inspection → completion
A simple record of completed activities might count all events without revealing the loop.
But repeated movement back to an earlier activity is operationally significant.
Rework loops can increase:
- cycle time;
- resource use;
- work in progress;
- material consumption; and
- production variability.
Sequence preserving records make these loops visible.
18. Rework Should Not Automatically Be Attributed to Sequence
If rework increases after a sequence change, the relationship deserves investigation.
But other explanations remain possible.
Input quality may have declined.
Inspection criteria may have become stricter.
Measurement may have improved.
Equipment may have changed.
Environmental conditions may affect production quality.
The evidence may establish that rework increased after a sequence change.
Stronger attribution requires additional support for a mechanism connecting the changed sequence with the observed defect or rework pattern.
19. Interruptions Alter the Effective Sequence
The documented operating procedure may specify:
A → B → C → D
Actual production may be:
A → B → interruption → B restart → C → D
The formal sequence appears unchanged.
The effective production history is different.
Interruptions can change:
- material condition;
- equipment state;
- operator context;
- waiting time;
- setup requirements; and
- downstream timing.
Production records should therefore distinguish planned workflow from observed workflow.
20. Planned Sequence and Actual Sequence Are Different Evidence
A process document states what should happen.
An operating record can show what did happen.
Both are useful.
They answer different questions.
The planned process provides the intended structure.
The production record provides evidence of execution.
Comparing the two can reveal:
- deviations;
- skipped activities;
- unexpected loops;
- delayed handoffs;
- additional work; and
- alternative operating paths.
Researchers should not substitute the planned process for observed production history.
21. Sequence Change Can Alter Resource Conflicts
Two activities may require the same worker, machine, tool, space, or supporting system.
Changing their order can create or remove conflicts.
A resource may become unavailable exactly when another activity requires it.
This can generate waiting even when total resource capacity has not changed.
The relevant production issue is therefore not only:
How much resource exists?
but:
When is that resource required by each activity?
Sequence determines part of that temporal relationship.
22. Handoffs Are Part of the Production Sequence
Production often moves between:
- people;
- machines;
- departments;
- locations; or
- operating stages.
Each transition can create delay or information loss.
A sequence change may introduce an additional handoff.
It may remove one.
It may move a handoff to a different point.
Records that preserve only processing activities can miss this change.
Where handoffs materially affect production, they belong in the observed process history.
23. Batching Can Change the Meaning of Sequence
Some work is processed individually.
Other work is grouped into batches.
A process may change from:
produce → inspect → release each unit
to:
produce many units → inspect batch → release batch.
The activities remain similar.
The sequence structure changes.
Batching can affect:
- waiting;
- inventory;
- inspection timing;
- feedback speed; and
- the number of units affected before a problem is detected.
Production records should therefore preserve batch boundaries where they materially affect process behavior.
24. Feedback Delay Can Amplify Sequence Effects
Suppose inspection occurs immediately after production.
A defect can be detected quickly.
Now move inspection to the end of a large batch.
The same inspection still occurs.
But feedback arrives later.
More units may be produced before the problem becomes visible.
The effect of sequence is therefore not only the physical order of tasks.
It can also change how quickly the system learns from its own output.
This is particularly important where downstream information influences upstream operation.
25. Sequence Can Affect Material State
Some production activities change the physical or biological state of materials.
Waiting between activities may therefore matter.
Material can:
- cool;
- heat;
- dry;
- absorb moisture;
- degrade;
- settle;
- harden;
- ferment;
- oxidize; or
- otherwise change.
In these systems:
A → immediate B
may not be operationally equivalent to:
A → long wait → B
Activity order alone is insufficient.
The interval between activities becomes part of the sequence evidence.
26. Sequence and Timing Must Be Separated
Two production units may follow the same order:
A → B → C
Unit 1:
A → 5 minute wait → B → 5 minute wait → C
Unit 2:
A → 3 hour wait → B → 2 hour wait → C
Sequence is identical.
Timing is not.
If outcomes differ, the evidence does not justify calling this purely a sequence effect.
Production research should distinguish:
order effect
from
interval effect.
The two can interact, but they are not identical.
27. Sequence and Workload Must Also Be Separated
Suppose a production sequence changes during the busiest month of the year.
Queues increase.
Lead time rises.
Output becomes more variable.
The new sequence may be responsible.
Higher workload may be responsible.
Both may contribute.
Without workload records, attribution remains uncertain.
Researchers should therefore consider production demand or operating load when workload could affect flow.
28. Equipment Condition Can Confound Sequence Comparisons
A sequence introduced after major maintenance may appear more efficient.
Another sequence used during equipment deterioration may appear worse.
The observed performance difference contains at least two changes:
- sequence; and
- equipment condition.
Maintenance and failure evidence can therefore provide important context when comparing work sequences through time.
29. Environmental Conditions Can Confound Sequence Comparisons
Production sequence may also interact with the operating environment.
A waiting period that is harmless under one temperature or humidity may affect materials under another.
A work order suitable during dry conditions may behave differently during wet conditions.
Environmental context can therefore alter the consequences of the same sequence.
Sequence research should not assume environmental equivalence when the production mechanism is environmentally sensitive.
30. Sequence Changes Can Be Gradual
Operational change does not always occur on one clearly documented date.
Workers may begin using an alternative path occasionally.
The alternative becomes more common.
Eventually it becomes the dominant sequence.
If researchers compare only two dates, they may miss the transition.
Repeated records can reveal changing sequence frequency:
10 percent alternative path;
25 percent;
50 percent;
80 percent;
then nearly all production.
This allows operational change to be studied as a process rather than a single event.
31. Sequence Stability Is Itself a Production Characteristic
A system may follow one highly consistent path.
Another may contain many sequence variants.
Neither is automatically better.
Flexible systems may require multiple valid paths.
Highly variable sequences may also indicate:
- exceptions;
- unclear procedures;
- resource constraints;
- rework; or
- unstable operation.
The number and frequency of process variants can therefore provide evidence about production organization, but interpretation depends on system design.
32. Historical Records Can Reveal When a Sequence Became Normal
Longitudinal records can establish that an operating path once rare later became routine.
This can matter when interpreting historical output.
A production period from several years earlier may have occurred under a substantially different workflow even if:
- the equipment is still present;
- the same activities still exist; and
- the final product is unchanged.
Historical production comparison should therefore consider process continuity as well as asset continuity.
33. Missing Events Can Distort Reconstructed Sequences
Suppose an activity actually follows:
A → B → C → D
but B was not recorded.
The reconstructed sequence becomes:
A → C → D
Researchers may incorrectly infer that B was skipped.
This is a critical limitation of event based production research.
Absence from the record does not automatically establish absence from the process.
The completeness of event capture determines how confidently observed sequences can be interpreted.
34. Increased Recording Can Create Artificial Process Complexity
A new digital system may begin recording activities that were previously invisible.
Historical records show:
A → C
New records show:
A → B1 → B2 → C
The production process may not have changed at all.
Observation became more detailed.
Without considering the history of the recording system, researchers could incorrectly conclude that additional production stages were introduced.
Changes in observation must therefore be separated from changes in operation.
35. Event Definitions Must Remain Comparable
An event labeled “completed” may change meaning over time.
One period may define completion as:
physical processing finished.
Another may define it as:
processing finished and quality verified.
The same event label then represents different process states.
Sequence analysis depends on semantic consistency as well as timestamps.
Researchers should consider what each recorded event represented during the period being compared.
36. Process Mining Provides a Useful Analytical Perspective
Process mining research demonstrates how event logs can be used to discover, compare, and evaluate actual process behavior.
A useful event log generally connects:
- a process instance or case;
- an activity; and
- a timestamp.
Additional attributes can provide information about:
- resources;
- locations;
- cost;
- quality; or
- operating conditions.
This allows researchers to reconstruct process variants from recorded events rather than relying only on intended procedures.
The principle is valuable for production research even when a full process mining system is not used.
Structured event history can preserve analytical possibilities that aggregate production totals cannot.
37. Conformance Can Reveal Deviation From the Intended Sequence
When an intended production model exists, observed event sequences can be compared with it.
Researchers may identify:
- unexpected activities;
- missing recorded activities;
- different ordering;
- repeated steps; or
- alternative paths.
But deviation should not automatically be classified as failure.
Some deviations may represent:
- valid adaptation;
- exception handling;
- improved practice; or
- necessary response to changing conditions.
The record establishes deviation.
Its meaning requires investigation.
38. Sequence Performance Requires Multiple Measures
No single production measure captures all sequence effects.
Useful measures may include:
- throughput;
- cycle time;
- lead time;
- waiting time;
- work in progress;
- setup time;
- changeover frequency;
- rework;
- defect frequency;
- resource utilization;
- downtime;
- overtime; and
- output variability.
A sequence may improve one measure while worsening another.
For example, throughput may increase while work in progress also increases.
Lead time may fall while setup burden rises.
Researchers should therefore avoid declaring one sequence universally superior from a single indicator.
39. Before and After Comparison Is Useful but Limited
A simple investigation may compare:
before sequence change
with
after sequence change.
This can reveal meaningful differences.
But time itself introduces other possible changes.
Inputs change.
Equipment ages.
Workers learn.
Demand changes.
Maintenance occurs.
Environment changes.
Measurement improves.
The longer the interval between periods, the more opportunities exist for competing explanations.
Before and after evidence should therefore be interpreted as longitudinal comparison, not automatically as controlled experimentation.
40. Repeated Switching Can Strengthen Evidence
Some systems alternate between different valid sequences.
If performance differences repeatedly appear when sequence changes and reverse when the original sequence returns, the relationship becomes more informative.
For example:
Sequence A → shorter waiting
Sequence B → longer waiting
Sequence A again → shorter waiting
Repeated correspondence can strengthen the case that sequence matters.
It still requires examination of competing variables, but it provides stronger evidence than a single historical transition.
41. Mechanistic Evidence Strengthens Attribution
Sequence attribution becomes stronger when researchers can explain how the changed order could produce the observed outcome.
For example, a sequence change may introduce an additional changeover, increasing setup and waiting and contributing to longer lead time.
Moving inspection later may delay feedback, allowing more units to be produced before a defect becomes visible and potentially increasing rework.
The mechanism connects process history with production outcome.
This is stronger than observing only that sequence and output changed during the same period.
42. Research Considerations for Comparing Work Sequences
Longitudinal sequence research involves several related but distinct forms of evidence.
Process identity affects sequence reconstruction.
Events that occur close together in time do not necessarily belong to the same production unit, batch, job, order, asset, or process instance. Sequence interpretation depends on knowing which events belong together.
Activity occurrence and activity order provide different information.
Two production periods can contain the same activities while organizing them differently.
Order and timing are not identical.
Two production units may follow the same activity order but experience very different waiting intervals. An observed performance difference should therefore not automatically be classified as an order effect.
Transitions can carry production consequences.
Setup, changeovers, handoffs, interruptions, and movement between activities can affect production even when the activities themselves remain unchanged.
Loops and alternative paths matter.
Rework, repeated activities, skipped recorded stages, and process variants can change cycle time, resource use, work in progress, and other outcomes.
Operating context affects comparability.
Workload, inputs, equipment condition, maintenance, staffing, environment, and resource availability may change at the same time as sequence.
Recording systems can alter the observed process.
Missing events can make a process appear shorter. More detailed recording can make an unchanged process appear more complex. Changing event definitions can also reduce historical comparability.
Performance is multidimensional.
Throughput, cycle time, lead time, waiting, setup, work in progress, quality, rework, downtime, overtime, and variability can move in different directions.
Mechanistic evidence strengthens interpretation.
A sequence difference becomes more informative when an appropriate mechanism such as additional setup, feedback delay, material state change, resource conflict, or bottleneck movement can connect it with the observed outcome.
Association remains distinct from causal attribution.
A sequence change and a performance change occurring during the same period establish co occurrence. Stronger causal claims require evidence sufficient to address relevant competing explanations.
These considerations do not constitute a universal method for evaluating production sequences.
They identify common evidentiary problems that arise when researchers use longitudinal production records to compare how work moved through a system.
43. What Production Records Can Establish
With sufficiently complete and comparable records, longitudinal evidence may establish that:
- activity order changed;
- particular sequence variants became more or less common;
- waiting changed;
- setup or changeovers changed;
- work in progress accumulated differently;
- rework loops increased;
- handoffs changed;
- bottlenecks emerged or moved;
- lead time changed;
- throughput changed; or
- the resources required to maintain the same output changed.
These findings are valuable even when causal attribution remains incomplete.
44. What Production Records Cannot Establish Alone
Production records generally cannot establish automatically:
- that sequence caused an observed output change;
- that the most frequent sequence is the best sequence;
- that a rare sequence is incorrect;
- that stable throughput means process performance is unchanged;
- that a missing activity was not performed;
- that a documented deviation represents operational failure; or
- that two periods are comparable merely because they contain the same activities.
Sequence evidence requires context.
45. Research Implications
Production records can preserve more than production quantity.
They can preserve production structure.
When identity, activity, order, timing, transitions, and outcomes are connected, researchers can investigate how work actually moved through the system.
This changes the analytical question.
Instead of asking only:
How much was produced?
researchers can ask:
Through what path was it produced?
How long did that path take?
Where did work wait?
Where did it return?
Which transitions required setup?
Where did work accumulate?
Which sequence variants were associated with different outcomes?
These questions expose operational changes that final production totals can hide.
46. Limitations
Detailed sequence research requires sufficiently granular production records.
Many operational systems preserve final output but not event history.
Historical records may contain missing events.
Timestamps may be inconsistent.
Different production units may be difficult to identify.
Event definitions may change.
Manual activities may remain undocumented.
Increased digitization can make recent processes appear more complex merely because observation improved.
Sequence may also interact with workload, equipment condition, input quality, maintenance, environmental conditions, staffing, and resource availability.
Observed associations should therefore remain separate from causal conclusions unless the evidence supports stronger attribution.
47. Research Conclusion
Production is not only a collection of activities.
It is a movement through activities.
That movement has order.
It has timing.
It has waiting.
It has transitions.
It can contain interruptions, loops, handoffs, bottlenecks, and alternative paths.
When production records preserve only final output, much of this operating history disappears.
When they preserve structured events through time, changing work sequences become visible.
Researchers can observe that the same activities occurred in a different order.
They can examine whether waiting changed.
They can identify additional setup.
They can observe accumulating work in progress.
They can detect rework loops.
They can see bottlenecks move.
And they can determine whether stable final output coincided with increasingly complex operational compensation.
These observations do not all support the same conclusion.
An observed change in activity order establishes a sequence difference.
A simultaneous change in waiting, setup, work in progress, rework, throughput, or another production measure establishes an additional performance difference.
Repeated correspondence across sequence variants can strengthen the relationship.
Evidence of an appropriate mechanism can strengthen interpretation further.
But a changed sequence followed by a changed outcome does not, by itself, establish causality.
This leads to a broader principle:
Production records become more informative when they preserve not only what work was completed, but how work moved from one state to the next.
The same activities do not necessarily create the same process.
The same resources do not necessarily create the same flow.
The same final output does not necessarily indicate the same operating performance.
Longitudinal production records make these distinctions researchable.
They allow researchers to examine how work actually occurred and to distinguish an unchanged list of activities from a production system whose observed organization changed through time.
References
van der Aalst, W. M. P. (2016). Process Mining: Data Science in Action. Second Edition. Springer. DOI: 10.1007/978-3-662-49851-4.
IEEE Task Force on Process Mining. (2012). Process Mining Manifesto. In Business Process Management Workshops. Lecture Notes in Business Information Processing, 99, 169–194. Springer. DOI: 10.1007/978-3-642-28108-2_19.
van der Aalst, W. M. P., Adriansyah, A., & van Dongen, B. (2012). Replaying history on process models for conformance checking and performance analysis. WIREs Data Mining and Knowledge Discovery, 2(2), 182–192. DOI: 10.1002/widm.1045.
Zhou, C., & Egbelu, P. J. (1989). Scheduling in a manufacturing shop with sequence-dependent setups. Robotics and Computer-Integrated Manufacturing, 5(1), 73–81. DOI: 10.1016/0736-5845(89)90031-8.
Vinod, V., & Sridharan, R. (2008). Scheduling a dynamic job shop production system with sequence-dependent setups: An experimental study. Robotics and Computer-Integrated Manufacturing, 24(3), 435–449. DOI: 10.1016/j.rcim.2007.05.001.
Mateus, G. R., Ravetti, M. G., de Souza, M. C., & Valeriano, T. M. (2010). Capacitated lot sizing and sequence dependent setup scheduling: An iterative approach for integration. Journal of Scheduling, 13, 245–259. DOI: 10.1007/s10951-009-0156-2.
Little, J. D. C. (1961). A Proof for the Queuing Formula: L = λW. Operations Research, 9(3), 383–387. DOI: 10.1287/opre.9.3.383.