A chart shows what happened. An annotated chart explains it. The gap between those two is where most of the value of a visualization lives, and it is the part most often skipped because annotation feels like commentary rather than data.

It is not commentary. An annotation is the answer to the question the viewer was about to ask, placed where they were about to ask it.

Consider a line chart showing a metric that dropped sharply in March and recovered by May. Everything a viewer needs to see the drop is present. Nothing tells them whether it was a data collection outage, a pricing change, a seasonal pattern that repeats every year, or a genuine crisis.

Without that context, each viewer constructs their own explanation, and those explanations will differ. Someone will conclude the product is failing. Someone else will remember the migration and dismiss it. Both are reading the same chart. The chart has not communicated; it has invited speculation.

An annotation reading "collection outage, 12-19 March" resolves it in four words. That is a better return on space than any additional series.

There are a handful of annotation types, and each has a specific job.

Reference lines mark a threshold: a target, a limit, a break-even point, a service level objective. They convert an absolute reading into a judgement, because a value is only good or bad relative to something.

Reference bands mark a range rather than a line: an acceptable operating window, a confidence interval, a forecast range. Bands are particularly valuable for signalling uncertainty, which a single line actively hides.

Event markers mark a moment on a time axis: a release, a campaign launch, a policy change, an outage. They answer why the line moved, which is the most common question anyone asks of a time series.

Region shading marks a period: a promotional window, a period of missing data, a recession, a pre-launch phase. Shading is better than two boundary lines because it reads as a span rather than as two separate events.

Point callouts mark a specific observation: the peak, the outlier, the current value. They are the most direct form of annotation and should be used sparingly, since a chart with eight callouts has none.

Direct labels replace a legend by putting each series name next to the series itself.

A legend forces a round trip. The viewer looks at a line, moves to the legend, matches a color, holds the name in memory, and returns. With four series that is four round trips, repeated every time attention shifts.

Placing the series name at the end of each line eliminates the trip entirely. The name is where the eye already is. This single change often does more for readability than any other adjustment, and it frees the space the legend occupied.

The technique has limits. Lines that converge at the right edge leave no room, in which case label at the point of maximum separation instead. Many series make labels collide, which is a signal that you have too many series rather than a labelling problem. On narrow screens direct labels may not fit, and a caption naming the series in reading order can substitute.

Event markers are the highest-value annotation in dashboards, and the most neglected, because they require information that does not live in the metrics database.

The practical approach is to maintain a small list of significant events with dates and short descriptions, and render them as light vertical lines with compact labels. Keep the marker visually subordinate to the data, since it is context rather than the subject. Where events cluster, group them behind a single marker rather than drawing six overlapping lines.

The maintenance burden is real but small, and the payoff compounds. Six months later, nobody remembers what happened in March, and the marker is the only reason the chart is still interpretable.

Annotation text is read in a glance, under a chart, by someone whose attention is on the data. It should be short, specific, and free of hedging.

State what happened, not what the chart shows. "Pricing change" beats "note the decline here," which tells the viewer something they can already see. Include the date if it is not obvious from position. Avoid internal jargon and project codenames, since annotations outlive the people who understood them.

Where an annotation needs more than a dozen words, it belongs in a caption or a note below the chart rather than inside the plot area. Long text inside a chart competes with the data for the same space and loses.

Annotations should be clearly present and clearly secondary. That means lighter weight than the data: thinner lines, dashed rather than solid, muted color, smaller text. A reference line drawn as heavily as the series competes with it, and viewers may misread the threshold as data.

Avoid placing annotations over data. Position them in whitespace, and use a short leader line to connect the label to the point it describes. Where the plot area is crowded, moving the annotation outside the axes and pointing inward is usually better than overlapping the series.

Consistency matters here as much as anywhere. If targets are dashed grey lines in one chart, they are dashed grey lines in every chart, so viewers learn the vocabulary once.

Annotation can be overused, in three specific ways.

Annotating the obvious adds clutter without information. If a chart has one dramatic spike, a callout reading "spike" is noise.

Annotating everything defeats the purpose. Annotation works by contrast: a marked point stands out because the others are not marked. Mark every point and nothing stands out.

Annotating instead of fixing the chart is the subtlest failure. A note explaining that the axis does not start at zero, or that two series use different scales, is a patch over a design problem. Fix the chart.

There is a longer-term argument for annotation that has nothing to do with today's reader.

A dashboard is consulted for years. The people who understood why the line moved leave. The context that was obvious in the moment becomes unrecoverable. An annotated chart carries its own explanation forward; an unannotated one becomes a set of shapes that nobody can interpret, and eventually somebody rediscovers a pattern that was explained and resolved three years earlier.

Annotations are the cheapest documentation you will ever write, and the only documentation that is guaranteed to be read, because it lives inside the thing people are already looking at.