Design Time Ranges Around Stored Resolution

Data & Dashboards · charts, dashboards, metrics

Updated 2026-08-04

Understand that historical data gets coarser with age. Systems that record a point every second keep that detail briefly, then downsample it into minutes, then hours, so a year-old view cannot show a spike that lasted ten seconds.
Say which resolution the chart is drawn at. Hourly averages on the axis explains why last week looks smooth and today looks jagged.
Never let a range change silently alter the shape of the answer. A spike visible at one zoom level and absent at another is the most confusing thing a chart can do.
Match the aggregation to the question. An average hides the spike that caused the outage, a maximum shows it, and a dashboard about reliability should not be using the average by default.
Offer ranges that align with what is actually stored. A one second granularity option over ninety days is a control that will either fail or lie.
Show where the data ends rather than drawing zero. A missing period rendered as a flat line at zero reads as an outage that did not happen.
Mark the incomplete current interval. The bar for this hour is smaller because the hour is not finished, and users will read it as a drop.
Keep retention visible in the interface. If detail is only kept for thirty days, that fact belongs beside the range picker and not in a support article.

A time series chart is a summary whose resolution changes with age, and the axis is where the interface admits it.