Backlog composition and cycle-time interpretation
An aggregate cycle-time figure can change because the work mix changed. A period with more simple requests may look faster even if every class stayed the same. A period that closes only easy items can look faster while older open work remains unresolved.
This study asks two separate questions: did the composition of eligible work change, and did time change within stable work classes? Keeping those questions separate prevents a mix shift from being presented as process improvement.
Freeze arrival cohorts and work classes
Define one arrival event, one completion event, and a fixed intake window. Retain completed, open, cancelled, duplicate, and indeterminate items. Classify work using fields known at intake so the final outcome does not influence the class.
Use a small set of operationally meaningful classes such as request type, service path, evidence requirement, or approved complexity band. The NIST process handbook starts characterization with defined inputs, outputs, and behavior, which supports stable class definitions before comparison.
Publish a composition table for each cohort. Show item counts and shares by class, source, priority, and disposition. If classification is missing, retain an unknown class rather than assigning records to the dominant group.
Keep unfinished time visible
Cycle time is not yet observed for an open item. Excluding all open work selects on completion and can create a favorable picture. NIST describes right censoring as cases where the event time lies beyond the observation boundary, a concept that applies when the study ends before work closes.
At minimum, show the age and count of open items beside completed-item cycle time. For a formal time-to-event analysis, choose methods suited to censoring and obtain statistical review. Do not replace open duration with zero or a guessed completion date.
Report medians and selected percentiles for stable classes with adequate evidence. Provide counts so readers can see sparse groups. A weighted standardized comparison can answer what each period would look like under one common mix, but the weights and reference mix must be frozen before results are interpreted.
Separate mix shift from within-class movement
First compare class shares between cohorts. Next compare duration distributions within each class. Finally, show the aggregate result. If the aggregate moved but class-specific results did not, describe the evidence as composition change rather than a speed change.
Check due-date rules, clock definitions, and extraction dates for drift. The GAO reliability guide emphasizes whether data are applicable to the intended question, so a clean export can still be unsuitable when definitions changed.
An operations coordinator can preserve cohorts, manage classification rules, and publish coverage and censoring. The process owner approves class boundaries and any causal investigation. Our operations support services can maintain this evidence, and the Research index offers related flow studies.
Sources and limits
- NIST process characterization, for defined process inputs and outputs.
- NIST censoring guidance, for unfinished event times at an observation boundary.
- NIST distribution exploration, for examining empirical distributions.
- NIST histogram guidance, for showing distribution shape rather than only averages.
- NIST sample selection, for representative process sampling.
- NIST change detection, for distinguishing process change from routine variation.
- GAO reliability guide, for accuracy, completeness, and intended use.
- Census analysis standard, for supported conclusions and stated uncertainty.
- Census administrative records, for evaluating coverage and record quality.
- BLS seasonal methodology, for composition, calendar, outlier, and revision considerations in time series.
These sources support cohort, distribution, and censoring controls. They do not prove that backlog mix caused an outcome or prescribe a universal work classification. This article is operational research guidance, not statistical, audit, staffing, or service-level advice.