A simulation starting empty can measure a startup episode before it resembles its intended continuing operation. Keep that episode visible when choosing an observation window.
Separate the initial state from the target
Output analysis for a nonterminating simulation must consider initialization bias when its convenient starting state differs from the intended long-run operation. A warm-up interval can be excluded from the estimate, but its length needs justification. Removing an arbitrary first portion does not prove that the remainder is representative. Primary reference: Winter Simulation Conference: initialization bias and output analysis.
State whether the question concerns finite startup or continuing operation. If a shift intentionally starts empty, startup belongs to that target. The observation question determines whether an interval should be excluded.
Compare fictional observation windows
Suppose a hypothetical run records 2 completed jobs in its first 2 minutes and 40 in the following 8 minutes. Whole-run output is 42/10 = 4.2 jobs per minute. The latter window gives 40/8 = 5 jobs per minute. These are different defined summaries of the same invented ledger.
The arithmetic does not establish that minute 2 is a valid warm-up cutoff. Retain the time series and initial queue information while investigating whether the chosen later interval fits the intended target. Do not select a cutoff merely because it produces the preferred rate.
Review the declared analysis window
Prepare fields for initial conditions, target process, candidate warm-up interval, retained window and the reason for that choice. Preserve the startup data even when it is excluded from one summary, so the reported denominator and excluded output remain reviewable.
Further modeled runs and an appropriate output-analysis method may be necessary for the intended inference. This guide sets no universal warm-up duration, confidence method or HM throughput result. Its reader decision is whether a transient startup and a continuing-operation estimate have been silently combined.
Customer Questions
Is every startup interval unwanted bias?
No. It belongs in the result when startup is the actual question being modeled.
Does the fictional later-window rate prove stationarity?
No. It only summarizes the chosen eight-minute interval.
Can a cutoff be chosen solely to improve the result?
No. Its justification must relate to the intended observation target.
Primary References
These references support the technical principles discussed in this guide. The worked examples and review questions are educational.
Discuss Your Machine Requirement
Share your product, container, required output and the evidence needed for your technical review.
Request a Technical Review