LEARN. PRACTICE. GROW.Define the unit before counting
In this fictional lab, an item is a migrated file route accepted by its consumer. Creating a VM does not finish the route. Measurement window [10,15) includes acceptances on day 10 and excludes day 15. A and B, accepted on days 10 and 14, count; C, accepted on day 15, belongs to the next window; D, developed on day 13 without acceptance, remains open. Observed throughput is two. Use the same unit and boundary when comparing periods. A definition change should accompany the series so the committee does not mistake a measurement artifact for improvement.
Also inspect unfinished work
The last five completion times are 1, 2, 2, 3, and 4 days. The board retains two items aged 18 and 21 days. Do not conclude that the system always finishes within four days: that statement ignores work still inside it. Also do not add 18 and 21 to history as final durations, because they can increase. Show both perspectives and investigate the oldest item: an external decision, oversized work, or unvalidated integration. The purpose is to select a useful action rather than blame the person named on the card.
Build a small forecast you can explain
Only in this exercise, use weekly history [2,4,4,6]. The model independently samples one of four observations for each future week, with equal weight per observation. Both 4s remain because they occurred twice. For two weeks, enumerate 16 ordered pairs. Five fall below eight: 2+2, two occurrences of 2+4, and two of 4+2. Eleven reach at least eight, producing 68.75% in the model. This enumeration teaches distribution effects; it is neither a calibrated forecast for a real team nor an algorithm mandated by PMI or Kanban.
Use probability for an explicit decision
The fictional sponsor asks for at least 85% within the model before communicating a commitment. Eight items in two weeks do not satisfy that rule. Six items reach the threshold in 15 of 16 pairs, or 93.75%, but only under the same assumptions. Do not present scope reduction as an automatic recommendation: confirm whether six routes constitute a useful delivery and whether dependencies permit the slice. A four-week sample is small; correlation between weeks, size variation, holidays, and RUN changes can make the model unsuitable. Disclose those limits alongside the number.
Reserve capacity without counting a person twice
The plan has 20 person-days after absences. It reserves 12 for RUN and three for training and ceremonies without overlap, leaving five. An eight-person-day project proposal exceeds the balance by three. This arithmetic exposes a mismatch; it does not automatically predict dates or incident distribution. If support grows, review the reservation and delivery options. A role described as 75% support provides work context rather than a rule for every team or proof that the time is removable repetition. Distinguish novel investigation, improvement, and recurring manual work.
Separate scope growth from progress
Scope increased from 20 to 25 items and acceptances from ten to 14. Items have equal weight in this exercise, so the state is 14/25=56%. The team delivered four and demand grew by five; both changes matter. A burn-up with scope and completion lines supports discussing that difference. Do not retain the old denominator to announce 70% or treat arrivals as deliveries. If item sizes or values differ substantially, count-based percentage needs additional context and does not automatically represent the percentage of realized benefit.
Guided practice: prepare the committee
Recalculate throughput of two, probabilities 11/16 and 15/16, the five-person-day balance, and the burn-up’s 56%. Then write a four-line update: observed result; changed assumptions; scope or date options; required decision and owner. State what the model does not prove: neither guaranteed migration approval, individual capacity, nor actual week-to-week independence. For an authorization-blocked item, add who can decide and when they will be contacted. Update quality depends on the decision it enables rather than the number of charts.
Synthetic exercise; no production data or calibrated forecast.
weekly_observations = [2, 4, 4, 6]
ordered_two_week_pairs = 4 * 4
pairs_at_least_8 = 11 # 68.75% under independent resampling
pairs_at_least_6 = 15 # 93.75% under the same assumptions
project_person_days = 20 - 12 - 3 # 5
current_scope = 20 + 5
accepted = 10 + 4
accepted_percent = 100 * 14 / 25 # 56%Eight routes have 68.75% in the two-week model; six have 93.75%. Increased RUN demand requires reviewing comparability before using either figure.
Common pitfalls
Excluding slow weeks, hiding blockers, confusing mean with guarantee, counting development as acceptance, or allocating the same availability to RUN and project work.
Related topics: Flow metrics · Prioritization and learning
A forecast is conditional. Accompany the number with its unit, window, assumptions, uncertainty, and required decision.
Reference: The Kanban Guide · PMI-ACP ECO March 2026