Insights Data
Do not interpolate: draw the holes in your data
An interpolated line looks perfectly healthy, and that is exactly the problem. We made the dashboard uglier, and it caught a six-hour outage.
Interpolation is a considerate lie
A segment of data is missing. What should the chart draw? Most visualisation libraries default to connecting across the gap. The resulting line looks smooth, healthy, unremarkable — and the six hours it describes contained nothing at all, because there was no data. The tool is not malicious. It simply assumes that looking complete matters more than being truthful.
Some systems have gaps by nature
The ten thousand Verde sites sit on hillsides, on rooftops, on connections that drop. Data arrives late, arrives twice, or does not arrive. This is not an anomaly, it is the normal condition. A dashboard that assumes completeness lies to its readers every day on a system like this, and it lies persuasively.
Draw the gap as a texture
Our approach was unremarkable: where data is missing we draw no line and lay down diagonal hatching instead. It is conspicuous, and it cannot be misread as a low value — which is the other common mistake, filling gaps with zero, so that an interval with no data looks exactly like a total outage.
The client chose the ugly version
The curves in the first design comps were continuous and handsome. We built both versions against real data, put them side by side, and explained the circumstances under which each would mislead them. They chose the hatched one. We did not have to argue: the two charts made the argument.
Six months later it earned its keep
A region stopped reporting for six hours. Because a band of hatching appeared on the chart, operations noticed and acted the same day. In the interpolated version those six hours would have been a slightly declining but entirely plausible line, and the outage would have surfaced at month-end reconciliation — assuming anyone went looking.
Where the principle stops
Not every gap deserves to be drawn. Three missing samples in a per-second feed render as noise and nothing else. The test is whether this gap, if it represents a failure, is something the reader needs to know about. At Verde it was. On a web traffic chart it usually is not. A principle without a stated boundary becomes dogma.