Open-work cycle-time right censoring
A completion-only cycle-time report can make a newer or slower cohort look artificially fast because the hardest items may still be open. Keep every item that entered the cohort, distinguish observed completions from unfinished follow-up, and show how the conclusion changes under more than one defensible summary.
This design studies reporting bias, not future completion. It does not assign a finish date to an open item or claim that every long-running item will become late.
A concrete NIST example shows why open observations cannot be discarded: its sample has 20 units, 10 recorded failures from 54 through 386 hours, and 10 units censored at 500 hours. The NIST Weibull example retains all 20 records when estimating the survival distribution.
Freeze entry cohorts and clocks
Define the entry event, completion event, eligible population, observation cutoff, and clock before calculating duration. Preserve the original time zone and timestamp precision, then record which items have a fully observed completion at the cutoff.
Use the same entry rule for every comparison group. A queue opened in one period and completed in another still belongs to its original entry cohort when the question concerns work that arrived during that period.
Separate calendar duration from any approved business-time duration. A calendar definition can change exposure, so retain its version and effective dates rather than silently replacing elapsed time.
Report unfinished exposure
For each cohort, show entered, completed, still open, withdrawn, and excluded counts. Report follow-up time for open items and the age distribution at the cutoff, because a cohort observed for ten days is not directly comparable with one observed for ninety days.
Publish the ordinary completion-only median beside a method that retains unfinished observations, such as a time-to-event curve when its assumptions and sample support are adequate. If that method is not appropriate, use transparent lower and upper descriptive bounds instead of hiding the open records.
Stratify by work type, priority, and entry period only when each stratum has enough records to interpret. Sparse strata should remain visible as counts rather than unstable percentages.
Test the conclusion for sensitivity
Repeat the comparison at several common follow-up windows. If the apparent winner changes when both cohorts receive equal follow-up, the original result depended on observation length rather than only operating performance.
Audit whether old open items were migrated, merged, or administratively closed near the cutoff. Those record changes can erase exposure without representing completed work.
Keep missing timestamps in a separate evidence class. An unknown completion time is not a completion and is not automatically an open item.
Keep decisions within the evidence
A censoring-aware study can show that completion-only summaries are incomplete. It cannot identify the cause of delay without further evidence about workload, dependencies, waiting, and data quality.
An operations coordinator can maintain cohort extracts, cutoff logs, and reconciliation records. The process owner decides which entry and completion events are authoritative and when specialist statistical review is warranted.
The backlog composition analysis addresses changing work mix, the response percentile analysis covers distribution summaries, and the Research index collects related designs.
Sources and limits
- NIST censoring overview, for the distinction between exact failures and right-censored observations.
- NIST Kaplan-Meier procedure, for product-limit estimates that retain censored records.
- NIST distribution selection, for checking censoring and recorded failure-time precision before modeling.
- NIST Weibull example, for the 20-unit worked example and its explicit censoring indicators.
- FDA time-to-event guidance, for defining events, censoring rules, and analysis populations prospectively.
- CDC life table methods, for official survival and life-table estimation practice.
- NIST reliability modeling, for selecting models that match the observed failure process.
- NIST reliability estimation, for estimating reliability functions from life-test data.
- STROBE cohort checklist, for reporting participant flow and follow-up clearly.
- OMB survey standards, for documented populations, exclusions, nonresponse, and reproducible analysis.
The numerical NIST example is instructional, not evidence about this site's workload. Use local entry, completion, and cutoff records to estimate local cycle time, and obtain statistical review before relying on a fitted survival model for consequential decisions.