§ 6.c.2 · charts

A chart is a comparison made visible.

Across 4,934 charts read by the corpus tagger, only one in four lands a strong takeaway. The rest decorate.

Corpus argumentative power n = 4,934 charts · 67 Ralph iterations

0% Absent

27% Weak

47% Medium

26% Strong

Absent literally no takeaway possible — should not have been a chart
Weak data shown, takeaway missing — decorative wallpaper
Medium the takeaway is there, but the reader has to work for it
Strong one comparison, made sharp — the reader leaves with a position they can repeat

02 the choice

Two charts. One choice.

Both ship data. Only one isolates a comparison the audience couldn't make in their head. Read each one cold; what do you walk away with?

Decorate data shown · comparison vague

A pair of nested-circle charts arguing market share

Form: nested proportional circles × 2
Comparison made: ambiguous — area encoding for ratios
Anti-pattern: dimensional vagueness · mismatched encoding

Reader leaves with: "Kakao operates in two markets" — but cannot remember which is bigger.

Argue one comparison · sharp · sourced

Grouped bar chart contrasting Arm royalties against peers

Form: grouped bar — entity × category
Comparison made: royalty multiple, two categories
Primitive: highlight palette · source line · numeric callouts

Reader leaves with: "Arm charges 3-5× the per-unit royalty of every peer."

03 six jobs

56 forms. Six jobs.

Bar, line, donut, waterfall — those are forms. Beneath the 56 canonical types, every chart is doing one of six jobs. Pick the job first; the form follows.

Evolve 31%

one thing changing over time

forms · line · area · bar-vertical (time series)

Compare 15%

A vs B at a single moment

forms · bar-grouped · bar-vertical · bubble

Decompose 15%

parts of a whole

forms · donut · pie · bar-stacked · waterfall

Rank 6%

order by magnitude

forms · bar-horizontal

Position 3%

where on a 2D landscape (x × y)

forms · scatter · bubble · matrix-2x2

Distribute 2%

shape of a distribution

forms · histogram · box-plot

04 anatomy

The chart, componentised.

The marks ARE the chart; five more primitives carry the editorial load; seven more sharpen it. Toggle Advanced to see the full anatomy — including four editorial concepts the corpus tagger does not yet detect.

1

Data marks

has_data_marks

the chart itself

The marks themselves — bars, lines, dots, slices. Without them there is no chart.

100% always present · captured as canonical_type across 57 named encodings, not as a boolean

Required when: Always. The marks ARE the chart. Removing them removes the chart, not a primitive.
Safe to skip when: Never. There is no such thing as an optional data block. If it is optional, you are looking at a layout, not a chart.

Full reference for has_data_marks →

2

Title

has_title

present

A chart-level title that anchors the takeaway in one phrase.

69.9% of corpus charts ship this

Required when: When the chart sits without a strong slide-level title carrying the takeaway, OR when the chart is shown out of context (in a report, in an embedded image).
Safe to skip when: When the slide title already states the takeaway in concrete terms.

Full reference for has_title →

3

Legend

has_legend

present

Mini swatches mapping colours/markers to series names.

30.5% of corpus charts ship this

Required when: Multi-series charts where colours encode different categories: stacked/grouped bars, multi-line, multi-area, donut-multi, choropleth.
Safe to skip when: Single-series chart, or charts where every element is labelled inline.

Full reference for has_legend →

4

Axis titles

has_axis_titles

present

Labels naming each axis (units, what the axis encodes).

14.5% of corpus charts ship this

Required when: Whenever the axis is a quantity that requires units (revenue, percentage, count). Even time axes benefit from a label when the period is non-obvious.
Safe to skip when: When the slide title implicitly states units ("Total revenue (\$M)") and the chart is in tight context.

Full reference for has_axis_titles →

5

Data labels

has_data_labels

present

Numeric values placed next to bars, points, or segments.

69.5% of corpus charts ship this

Required when: Hero charts where the takeaway depends on specific values. Sparse charts where each value is a deliberate beat (e.g., 5 bars in a hero chart).
Safe to skip when: Dense exploratory charts where the shape is the story (scatter, dense lines).

Full reference for has_data_labels →

6

Source note

has_source_note

present

Attribution to the underlying data ("Source: …").

39.2% of corpus charts ship this

Required when: Always — except for purely illustrative / schematic charts that explicitly say "Illustrative".
Safe to skip when: Schematic charts disclosed as illustrative.

Full reference for has_source_note →

7

Gridlines

has_gridlines

present

Reference lines across the plot area.

8.8% of corpus charts ship this

Required when: Dense data where the eye needs reference lines to read precise values (line charts with 30+ points, bar charts without data labels).
Safe to skip when: Sparse charts where data labels cover the precision need. Sparklines and small charts where gridlines crowd.

Full reference for has_gridlines →

8

Trendline

has_trendline

missing

A fitted line or annotation overlay calling out the rate of change.

0.4% of corpus charts ship this

Required when: When the rate / slope IS the takeaway. Common in growth-story bar charts and editorial scatters.
Safe to skip when: When the chart is purely descriptive (showing levels, not change).

Full reference for has_trendline →

9

Error bars

has_error_bars

missing

Visual representation of uncertainty around each point.

1.7% of corpus charts ship this

Required when: Forest plots, A/B test results, polling, any aggregate statistic with sample variance.
Safe to skip when: Census data, exact accounting figures, deterministic calculations.

Full reference for has_error_bars →

10

Subtitle

has_subtitle

not detected yet

A second line under the title that carries the metadata.

Editorial concept · not yet a corpus boolean. Roadmap target for the next enrichment pass.

Required when: When the title is an action title (a claim) and the reader still needs to know n, period, or units to verify.
Safe to skip when: When the data is universally legible (e.g. "Apple revenue, $B, 2010-2024" already in the title) or context lives in the slide title above.

Full reference for has_subtitle →

11

Annotation

has_annotation

not detected yet

A floating callout pointing at the data — explains what the eye should grab.

Editorial concept · not yet a corpus boolean. Roadmap target for the next enrichment pass.

Required when: When the takeaway hinges on one specific point, region, or change — and the reader would otherwise have to infer it.
Safe to skip when: When the shape itself argues (a clean exponential, a sharp spike) and the reader cannot miss it.

Full reference for has_annotation →

12

Highlight

has_highlight

not detected yet

One mark accented; the rest greyed out — visual subordination.

Editorial concept · not yet a corpus boolean. Roadmap target for the next enrichment pass.

Required when: When one item among many is the protagonist — the chart's job is to compare ONE against the rest.
Safe to skip when: When all categories must read equally — a true comparison with no protagonist.

Full reference for has_highlight →

13

Reference line

has_reference_line

not detected yet

A horizontal benchmark — average, target, or threshold.

Editorial concept · not yet a corpus boolean. Roadmap target for the next enrichment pass.

Required when: When the takeaway is "above/below the bar" — a target hit, a peer benchmark, an industry average.
Safe to skip when: When the data has no external reference, or the comparison is internal (this year vs last year).

Full reference for has_reference_line →

05 catalog

Pick the job. The shape follows.

56 canonical types, organised by the job each one is hired to do. Top three forms per job; the full reference lives at /insights/charts/types.

Evolve

how did it change over time?

30.5%

bar-vertical bar
bar-stacked-vertical bar
column-vertical bar

Compare

A vs B at one moment?

14.8%

bar-vertical bar
bar-horizontal bar
bar-stacked-vertical bar

Decompose

parts of a whole?

14.6%

bar-stacked-vertical bar
bar-stacked-horizontal bar
area-stacked time

Rank

who is biggest, then next?

5.6%

bar-vertical bar
bar-horizontal bar
bar-grouped-horizontal bar

Position

where on a 2D landscape?

2.5%

bubble corr
matrix-2x2 spec
venn spec

Distribute

shape of the distribution?

2.3%

bar-range bar
pie comp
donut comp

Forecast

projected vs known?

1.8%

line time
scatter-with-trendlines corr

Flow

what moves where?

0.2%

sankey flow
flow-diagram flow
funnel flow

System

how do the parts connect?

1%

flow-diagram flow
gantt flow
timeline-chart flow