On a cumulative flow diagram, what do the width and the slope of a band tell you?

## What the chart actually plots A cumulative flow diagram stacks running totals. For every date on the horizontal axis it plots, for each stage of the workflow, the cumulative number of items that have *reached that stage or gone past it*. The topmost curve is everything that has arrived; the bottom curve is everything that has been delivered. Because the totals only ever accumulate, no curve can fall — a curve that dips means items were removed from the board, which is a data problem worth chasing on its own. The coloured area between two adjacent curves is a **band**, and each band belongs to one stage. Nearly everything the chart tells you is read off the geometry of those bands rather than off any single plotted point. ## Reading the geometry | Geometry | What it means | What it usually signals | |---|---|---| | Vertical thickness of a band | items currently in that stage — work in progress | a thick band is a lot of unfinished work parked in one place | | Slope of a curve | the rate items crossed that boundary — throughput | a flattening curve means completions have stalled there | | Horizontal gap between top and bottom curve | roughly how long an item took end to end | a widening gap means the wait is growing | | A band widening over time | arrivals into the stage outrunning departures | a queue is accumulating in front of that stage | | A band narrowing | departures outrunning arrivals | the stage is draining, whether by finishing or by starving | | A flat band top | nothing entering the stage | upstream has stopped feeding it | The three primary reads are worth memorising in that order: **thickness is work in progress, slope is throughput, horizontal distance is time in the system.** An interviewer showing you a chart is almost always asking you to name one of those three and say what it implies. ## A worked read Over one 3-week window on a construction-site safety product, the band for the verification stage grew from 3 items wide to 17 items wide across 11 days, while the band for active development held steady at about 6. The bottom curve — delivered items — flattened noticeably over the same stretch. That composite tells a specific story. Development kept starting and passing on work at its usual rate, so its own band stayed constant. Verification accepted that work but stopped passing it out, so its band spread. Delivery flattened because delivery is downstream of verification. The constraint is located at verification, and the widening band, not the flat delivery line, is the piece of evidence that locates it — the flat delivery line only says something is wrong somewhere. The same shape has more than one cause, and the chart does not choose between them: a reviewer may be absent, the stage may depend on a shared environment, or the items may be arriving in a state that fails verification. The chart narrows the search to one stage; it does not name the reason. ## Honest limitations - **It is a lagging, aggregate picture.** Every point is a running total, so a fresh problem takes days to become visible as a shape. - **It says nothing about any individual item.** Band thickness is a count, and the horizontal gap is an approximation across the population. You cannot point at a dot and say which item it is, because there are no dots. - **It hides age.** A band of 17 items looks identical whether those items arrived yesterday or have been sitting for a month. Detecting stalled individual work needs a different chart. - **It is sensitive to how the board is drawn.** A stage split into two columns splits its band; a queue column merged into an active column hides the queue inside a thicker band. - **Curves that fall are a warning about the data, not about the flow.** Items deleted, restarted or moved backwards distort the totals, and the chart cannot distinguish those from real progress. ## What to say when handed one 1. Orient first: name the axes, identify the top and bottom curves, and state what a band means before interpreting anything. 2. Read thickness and slope separately, out loud, rather than jumping to a conclusion about the team. 3. Locate the widest and the fastest-widening band and say why that is where the queue is. 4. Say explicitly what the chart cannot tell you — which item, whose fault, and how old the oldest work is. 5. Treat it as a prompt for the next question rather than an answer in itself; the value is that it points at one stage instead of the whole workflow.