Engine 19.0.0rc597 · Reader 9.0.0rc656 · Development snapshot
What does the number measure?
Theil describes how events are distributed across people. Lead time follows a work object from its first recorded observation to closure. A survival curve includes objects that are still open. These measures answer different questions, even when they use the same event log.
Choose a measure below for its formula, a worked example, the published fields and its use in the Reader. Each article explains which records enter the calculation and when the Engine withholds a result.
Statistical measures
- Theil index
- How concentrated are process events across people? Calculate the index, individual contributions and the expected imbalance from random allocation.
- Kernel density estimation (KDE)
- What does a distribution of processing times look like? Read the log-time density, bandwidth and comparison curves.
- Statistical process control (SPC)
- Which processing times exceed the calculated control limits? Follow the log transformation and the moving-range calculation.
- Median and percentiles
- How long do the middle and slower observations take? Compare P50, P90 and P95, with their different minimum sample sizes.
- Wilson confidence interval
- How uncertain is an observed proportion? See why two samples with the same percentage can support different conclusions.
- Relative risk and lift
- How does one segment's deviation share compare with the remaining population? Calculate the ratio and its Katz confidence interval.
Time and flow
- Lead time
- Elapsed time from a work object's first recorded observation to its recorded closure, with explicit treatment of open objects and incomplete histories.
- Processing time
- The duration of an activity and the sum of process-step durations for a whole object. Parallel steps can make the total exceed lead time.
- Waiting time
- Measured gaps between ordered steps and durations of declared external-wait activities. Read how overlaps affect the calculation.
- Work in progress (WIP)
- Daily stock based on first arrivals and final closures, alongside a time-weighted average of open intervals. Check the population used by each.
- Arrivals and closures
- Daily counts of objects entering and leaving a process, including how closure outcomes affect the result.
- Kaplan–Meier survival
- Estimate the share of objects still open at each age while retaining right-censored observations.
- Remaining time
- Conditional time-to-close estimates for an object that has already been open for a given duration.
- Due-date risk
- The estimated probability that an object remains open at its due date, and when the survival curve cannot support an estimate.
- SLA elapsed time
- Elapsed time under an SLA clock, with working calendars, pauses, targets and the source system's own result.
People and work
- Recorded work time
- Combine admitted work intervals without counting concurrent records twice. Inspect overlap and source agreement separately.
- Earned work time
- Weight attributed process events by their configured activity targets and compare the total with recorded work time.
Process paths
- Process variants
- Count work objects that follow the same ordered sequence of activities, with their share of the admitted population.
- Repeated steps and returns
- Distinguish immediate repetitions from a return to an earlier activity, and see what these counts can establish.
Before comparing two results
Start with the unit and the population. Events per person, objects per day and seconds per activity are different quantities. A daily open-object count can include records that cannot enter a survival estimate. Missing closure information can therefore change the denominator as well as the value.
Then check the clock. Elapsed duration, summed process-step durations, the union of a person's work intervals and time inside a configured working calendar are separate measurements. A weekend can increase lead time without adding any SLA working time.
A missing value needs its explanation. The Engine can withhold a statistic because the sample is too small, the event order is unknown or the required timestamps are unavailable. The article for each measure names its conditions. Treating a withheld value as zero changes the meaning of the result.
Start with the question, then select the statistic
For an uneven workload, start with Theil T and the counts behind it. For long durations, compare median and P90, then inspect the distribution. For a changing sequence of durations, use the control-chart definition.
A proportion and a comparison of proportions also differ. A Wilson interval describes uncertainty around one observed share. Relative risk compares two shares. A Kaplan–Meier estimate addresses time to closure while retaining eligible open objects. Choose the population before calculating any of them.
Version and scope
This measure reference follows Engine 19.0.0rc597 and Reader 9.0.0rc656. The supplied project identifies them as a development snapshot. Its published result contract contains 18 tables and 226 columns; the Reader has five workspaces: Overview, Processes, People, Objects and Timing.
The existing installation and command reference on this site covers Engine 18.6.0rc590 and Reader 8.11.0rc649. Follow the version stated on the page. The new Reader refuses older result files: regenerate them with the matching Engine before opening them.
The articles identify published fields and Reader views where applicable. A published field does not imply that every Reader view displays it. The worked examples use the calculations in the stated Engine version; external references explain the underlying statistical methods.