LEARN. PRACTICE. GROW.Define what enters and leaves
Before calculating a date, identify the population and boundaries. In a fictional fund service, a request enters execution when accepted by the team and leaves when APS validates delivery. Three requests waiting for that validation remain in progress. The same rule distinguishes elapsed time, active effort, and open-work age. If an item started on day 2 at 10:00 UTC and remains open on day 7 at the same time, it is five days old; two blocked days remain included. Also define the calendar, reopening treatment, and work unit. Splitting one request into two administrative tasks can double counts without doubling outcomes. A historical series is comparable only while these choices remain identified.
Read the distribution and open work
Four deliveries completed in 2, 3, 3, and 4 days have a three-day mean. This does not determine when two other items already open for nine and eleven days will finish. Show those ages separately with waiting causes and impact. For a sorted sample of 2, 2, 3, 4, 4, 5, 6, 8, 13, and 21 days, nearest-rank places P80 at the ceiling of 0.8 times ten: the eighth value is eight. The mean is 6.8. These describe the same sample differently, and neither guarantees an individual outcome. Explain the percentile method so another person can reproduce it; spreadsheets may use different interpolation. A favorable median can coexist with two old items blocking infrastructure retirement and prolonging dual operation.
Run a small model and explain its limits
The laboratory enumerates weeks with two, three, four, or five completions, equally likely and independent. Three weeks yield 64 sequences; twenty reach twelve completions, or 31.25%. Four weeks yield 221 successful sequences out of 256, about 86.3%. Enumeration is exact for this teaching model. It does not establish that a real team has that probability: there are only four possible values, capacity is treated as stable, and weeks are independent. An extended supplier outage may affect several successive weeks. More draws do not correct that assumption. If a specialist will be absent, discuss capacity scenarios before reusing the number. Do not reduce probability by an arbitrary constant; identify how the condition changes work that can finish.
Turn the forecast into a tracked decision
Completing the initial twelve items does not empty a queue still receiving three requests weekly. In a simple period, ten initial items plus fifteen arrivals minus eleven departures leave fourteen, without cancellations or splits. State whether the commitment concerns a fixed cohort, a continuous service, or a priority subset. Bring scope, completion conditions, confirmed capacity, dependencies, and a review point to the sponsor. For example: the forecast covers only identified requests, assumes specialist availability, and will be revisited when supplier confirmation arrives. A date can support a scope decision without being an unconditional promise. The exercise ends when you can explain which changed data would alter your recommendation, rather than merely reproduce the percentage.
from itertools import product
from fractions import Fraction
weeks = list(product((2, 3, 4, 5), repeat=4))
probability = Fraction(sum(sum(w) >= 12 for w in weeks), len(weeks))
# 221/256 under stipulated independent, equally likely weeks.
# This is not a measured probability for a real team.221 / 256 = 86.328125% within the model; reporting about 86.3% does not remove assumption uncertainty.
Common pitfalls
Completed-item mean as whole-portfolio timing; age as final duration; decimal precision as confidence; forgotten arrivals.
Related topics: Capacity and forecasting · Dependencies and outcomes
A useful forecast identifies population, boundary, assumptions, and the decision the next observation may change.
Reference: The Kanban Guide, May 2025 · Service Manual current public guidance; Kanban Guide May2025; DORA five-metric model; primary references reviewed 2026-09-30