Problem
The dashboard has eleven colours on it and all of them mean something. By the third week nobody sees any of them, and the one that indicates a genuine fault has the same visual weight as the brand-blue line that has been flat since launch.
Solution
Draw the normal state in greys. Reserve colour for abnormal, and make saturation carry severity.
This is the core of ISA-101 and the high-performance HMI school it comes from, developed for control rooms where an operator watches a screen for eight hours and has to notice the one thing that changed. Base graphics in greys and neutrals; roughly nine-tenths of the screen neutral; bright colour reserved for abnormal conditions and active alarms; muted or low-intensity colour for early-stage states and full saturation held back for the severe ones.
The consequence is what makes it worth adopting outside process control: anything coloured on the screen is by definition something wrong, so it gets found without being searched for. The operator is not scanning. They are waiting for colour to appear, which is a far cheaper cognitive task and one that survives fatigue.
The discipline is harder than the rule. Every team wants their series to be their colour, every chart library ships a categorical palette on by default, and brand guidelines arrive wanting the accent on everything. Each of those is individually reasonable and collectively fatal, because the budget being spent is attention, and it is fixed.
WCAG’s use-of-colour requirement pushes the same direction from a different angle: colour must not be the only means of conveying information. A page that is grey until something is wrong still needs the wrong thing to be labelled, positioned or shaped differently, not merely tinted.
Use when
The screen is monitored continuously rather than read occasionally, the normal state is the common state, and noticing change matters more than reading values. Control rooms, NOC walls, on-call dashboards.
Don’t use when
The page is analytical and every series is equally interesting. Exploration needs categorical distinction, and a grey exploratory chart is just hard to read. Comparison work wants a palette; monitoring wants a resting state.
Trade-offs
Grey pages look unfinished to stakeholders, and defending one takes political capital that gets spent repeatedly rather than once. Categorical distinction genuinely suffers: greyscale supports perhaps four distinguishable series before viewers give up on the legend. The scheme rests entirely on thresholds being right, because the alerting colour only appears when a rule fires, so a badly-set rule produces a page that is calm and wrong. And in a genuinely bad incident everything colours at once, which is exactly when the scheme provides the least discrimination.
Checklist
- What fraction of a healthy screen carries colour, and is it close to none?
- Does saturation encode severity, or is every abnormal state equally loud?
- Is anything coloured for a reason other than abnormality—branding, series identity, decoration?
- Does the abnormal state have a non-colour signal too, per WCAG?
- Which thresholds turn colour on, and when were they last reviewed?
- What does the page look like when six things are wrong at once?
- Do greys carry enough separation to still show structure and hierarchy?
- Does the scheme survive a projector, a cheap wall TV, and a colourblind viewer?
- Is there a categorical palette hiding in a chart library default somewhere on this page?
- Who has authority to say no to the next request for a coloured element?
Compare
Industrial HMI software built to ISA-101 is where the discipline is enforced rather than encouraged, because a control room has a regulator and a fatigued operator and neither tolerates decoration. Netdata goes the opposite way by default, colouring densely across hundreds of charts, which suits a tool used for investigation rather than for standing watch. Grafana ships categorical palettes on by default and puts the whole burden on the dashboard author, so the muted approach is achievable and almost never the path of least resistance. Datadog sits in between with monitor status driving colour on status-oriented views while general dashboards stay fully coloured, which is coherent per-view and inconsistent across a product someone uses all day.
Related
Semantic status color defines what the alerting colours mean once they appear. Dense small-multiple layout is the layout that most needs this discipline to work at all. Categorical series palette is the opposing requirement and the entry that covers when distinction wins. Process mimic is the display type this convention was written for. Wallboard mode is the other context where a calm resting state matters most.