Every chart is ranked by these principles.

Humans decode visual encodings with varying accuracy. Position on a common scale is the most accurate; color and area are the least. This hierarchy should guide every chart type decision: use bar charts (position) over pie charts (angle/area) when precision matters.

Position > Length > Angle > Area > Volume > Color

Certain visual properties (color, orientation, size, motion) are detected by the brain in under 200 milliseconds, before conscious attention kicks in. Use these sparingly to highlight what matters. When everything is highlighted, nothing is.

The eye automatically groups elements by proximity, similarity, enclosure, and connection. Elements that are close together are perceived as related. Use whitespace and alignment to create visual structure without adding lines or boxes.

The eye reads change in a line chart by judging slope. When most segments hover near 0° (too wide) or 90° (too tall), trends become hard to compare. Cleveland's banking rule says to choose an aspect ratio so the absolute slopes average around 45° - the angle at which differences in slope are easiest to discriminate.

The visual representation of numbers must be directly proportional to the quantities represented. Tufte's Lie Factor measures this: the size of the effect shown in the graphic divided by the size of the effect in the data. A lie factor of 1.0 is honest. Anything else distorts.

Bar charts and area charts must start at zero. Without a zero baseline, a 2% change can look like a 200% change. Line charts are exempt because they encode rate of change via slope, not magnitude via area.

Changes in a chart should reflect changes in the data, never changes in design. Inconsistent scales, shifted axes, or variable bin widths create false patterns that mislead even careful readers.

A dual Y-axis can read cleanly when the two metrics are causally related and on different units (temperature vs. ice cream sales, ad spend vs. signups). It misleads when the axes are scaled to make unrelated lines appear to track each other - any pair of trending series can be made to look correlated with the right scaling. Default to two stacked charts or an indexed-to-100 view when in doubt.

Maximize the share of ink devoted to data. Minimize gridlines, borders, backgrounds, and decoration. Ask of every element: does this help the reader understand the data? If not, remove it or reduce it.

Erase non-data-ink, within reason.

Repeat the same chart structure with different data across a grid. The consistent format lets the eye compare shapes and patterns instantly without learning a new layout for each panel.

The best design solution for a wide range of problems.

'How much?' → bar. 'When did it change?' → line. 'What share?' → treemap. 'How does it flow?' → Sankey. 'Who's winning?' → bump chart. The chart type should be chosen by the analytical question, not the data shape.

On a categorical axis the eye reads ranking before it reads labels. Alphabetical order forces the reader to scan every bar to find the biggest; value-sorted order makes the ranking visible instantly. Reserve alphabetical when the user genuinely needs to look up by name (e.g. a state directory).

Legends force the reader to look away from the data, decode a color, then look back - cognitive overhead that compounds with every glance. Whenever the chart has fewer than ~7 series, place the label directly next to its line, slice, or bar. Reserve legends for when direct labeling would create overlap.

"Revenue by Quarter" describes the chart; "Revenue grew 31% in Q3, led by Enterprise" describes the finding. Newspaper-style headlines that name the takeaway help readers anchor before they decode - and force the author to actually have a point. If you can't write the headline, the chart probably isn't telling a story yet.

Rainbow colormaps create visual boundaries where none exist in the data because human color perception is non-linear. Use sequential scales (viridis, inferno) for ordered data and diverging scales for data with a meaningful midpoint. Never use rainbow.

Hue is effective for distinguishing 5-8 categories but terrible for encoding magnitude. For quantities, use position or length. When color must encode a value, use luminance (light-to-dark) on a single hue, not multiple hues.

Roughly 8% of men and 0.5% of women have some form of color vision deficiency, most commonly red-green. Pick palettes that remain distinguishable in CVD simulators (ColorBrewer, Okabe-Ito, viridis). Pair color with a second channel - shape, position, label - whenever the distinction matters.

Every editorial and design decision should serve comprehension of the underlying data. Annotations, labels, and context should be placed directly on the graphic. If a reader needs to look at a legend, then cross-reference back to the chart, you've added unnecessary cognitive work.

Cherry-picking a timeframe, omitting comparison groups, or showing a metric without its base rate can completely reverse the conclusion a reader draws. Every graphic should include enough context for the reader to form their own interpretation.