Average work in progress: reconcile rate and elapsed time

Average work in progress is a time-weighted count of started but unfinished items, not a single board snapshot. Use this independent four-item exercise to save a reconciled record of average open work, completion rate and mean elapsed time, with explicit limits on what the numbers can establish.

Updated: 18 October 2026.

What does average work in progress actually measure?

A team can finish several jobs during a review period while carrying a different number at different moments. Counting completions answers how many items left the chosen workflow. Counting open items at one moment answers what was present then. Neither number, by itself, tells you the average open count across the whole observation window.

For this worksheet, an item becomes open at its supplied started point and leaves at its supplied finished point. Waiting between those points still belongs to its elapsed time. We do not measure hands-on effort, staff utilization or time spent exclusively in a waiting queue. Items not yet started are outside the chosen workflow boundary.

Your output will contain four complete item records, a time-segment ledger, three correctly labelled metrics and a dimensional cross-check. It will not recommend a staffing level, choose a work limit or promise a completion date. The mean and median exercise starts with supplied response durations. Here you reconstruct elapsed time and time-weighted open work from paired events.

Establish an empty-to-empty observation contract

Observe a fictional workflow from minute zero through minute eighty. There are no open items immediately before zero. Every item that enters during the window is listed below, and the last one finishes at eighty. There are no hidden items, cancellations or reopened records. These are supplied teaching conditions, not assumptions to import into a live business dashboard.

Item P starts at zero and finishes at thirty. Q starts at ten and finishes at forty. R starts at twenty and finishes at sixty. S starts at fifty and finishes at eighty. Times are elapsed minutes from one shared origin, not local clock timestamps. No time-zone, holiday or working-calendar conversion is involved.

Keep identifiers attached to both events. Subtract start from finish for each item: thirty, thirty, forty and thirty minutes. Their sum is 130 item-minutes. All inputs are fictional. No client records, product account, staff schedule or measured service outcome was used to construct this card.

Build the changing open-item count

List the unique event times in order: 0, 10, 20, 30, 40, 50, 60 and 80. Adjacent boundaries create seven segments. Within each segment, the set of open items stays constant. Include a start and exclude a finish when determining membership; process simultaneous changes together instead of manufacturing a momentary extra item or gap.

From zero to ten, P is open. Q joins for ten to twenty, then R joins for twenty to thirty. P finishes at thirty, Q at forty, and S joins at fifty. R finishes at sixty, leaving S until eighty. Write the identifiers before their count so another reviewer can reconstruct every row.

Multiply each count by its segment duration, then add the last column: 10+20+30+20+10+20+20=130 item-minutes. This matches the sum of individual elapsed durations. The two ledgers count the same supplied open-item time in different directions, which gives you a useful reconciliation before calculating any averages.

Reconcile the three metrics without mixing units

Average open work is 130 item-minutes divided by eighty minutes, or 13/8=1.625 items. Completion rate is four finished items divided by eighty minutes, or 1/20 item per minute. Mean elapsed time is 130 item-minutes divided by four items, or 32.5 minutes. Preserve the labels and denominators beside the results.

Now multiply the completion rate by mean elapsed time: (1/20 item per minute)×32.5 minutes=1.625 items. The minute units cancel. You obtain the same value as the time-weighted open count. This is an exact finite-record accounting check for our fully observed empty-to-empty cohort, not proof that a real operation is in a steady state.

The completion rate can also be written as three items per hour: eighty minutes is four thirds of an hour, and four divided by four thirds is three. Do not multiply three items per hour directly by 32.5 minutes and call the product an item count. Convert 32.5 minutes to 13/24 hour first; three times 13/24 is 13/8.

Why are a snapshot and a segment average different?

The maximum open count is three. A snapshot taken at minute twenty-five would also show three. Neither is the time average of 1.625. Keep peak load, observed snapshot and time-weighted average separate in a report. Otherwise a correct count from one instant can be placed under an incorrect average heading.

Taking the ordinary mean of the seven segment counts also fails here. Their sum is twelve, so twelve divided by seven is about 1.714. The last segment lasts twenty minutes while every other segment lasts ten. Equal weighting of unequal-duration segments answers a different question. The required numerator is the sum of count multiplied by duration, and its denominator is eighty minutes.

The shift-coverage worksheet also splits time, but compares available staff with an owner-supplied minimum. There is no staffing requirement in this card. Three open items do not establish three employees, three simultaneously active tasks or a shortage of anyone. A shared arithmetic primitive does not make the decision outputs interchangeable.

Where should this small check stop?

If an item began before the window or remains open after it, its complete elapsed duration is not the same as its contribution inside the window. Our simple reconciliation deliberately avoids that boundary problem. Do not drop the crossing item or invent a finish at the observation cutoff to make the totals match. Keep the extended calculation unresolved until the inclusion and boundary treatment are defined.

An unknown finish is not a zero-duration completion. Cancellation and reopening also need explicit recording rules before you count them as departures or new items. This fictional input contains neither. If your real list may omit work, arithmetic agreement between two incomplete copies does not prove that the source is complete.

Likewise, the relationship does not show that halving work in progress will automatically halve elapsed time while preserving throughput. Changing a workflow can change its rate and other conditions. We did not run such an intervention. Mean elapsed time is also not an individual delivery promise: the supplied durations include thirty and forty minutes, not four identical values of 32.5.

The critical-path example checks a different timing problem involving task dependencies. No predecessor network or project finish is calculated here. Keep the observed cohort, rate and average together rather than relabelling one as the duration of a new project.

Save the record and preserve its limits

On Windows, search Start for Notepad, open the application and create a new document headed “Open work reconciliation V1: fictional data.” Copy the four event pairs, four durations, seven segments and their item-minute products. Include both totals of 130, the three metric calculations and the explicit empty-to-empty observation contract.

V1 | window0..80min | empty before/after
P0..30 Q10..40 R20..60 S50..80
Item durations30,30,40,30; sum130item-min
Segment area130item-min; average130/80=13/8items
Completions4/80=1/20item/min; mean130/4=32.5min
Reconcile(1/20)×32.5=13/8items
Peak3 is not the average; no individual date promise.

Press Ctrl+S, choose Documents and save as open-work-reconciliation-V1.txt. Close the editor. In File Explorer, open Documents and double-click that file. Check that all seven segments, both 130 totals and the warning about a date promise remain present. The suggested fifteen-minute scope is a planning allowance, not measured participant performance.

MIT's opened course notes present the area-based reasoning for Little's law with an empty-to-empty simplifying example. The opened Kanban Guide distinguishes started-but-unfinished work, completion rate and elapsed started-to-finished time. These are narrow conceptual sources; our fictional events and arithmetic are independently constructed. No software simulation, live stability test or measured business improvement is claimed.

Anar's earlier published reflection describes delaying his own projects while pursuing perfection. Here the limited application is to finish one auditable record before designing a larger dashboard. It is not evidence of an operational experiment. The artificial intelligence starter guide offers broader tasks, while the version-reference exercise helps preserve which input a checked result belongs to.