Charts and data-visualisation engines

This page is the reference for the flowss engines that turn numbers into charts: Plot for plain XY charts and plotted formulas, the three chart grammars Vega-Lite, Vega and Plotly, and the specialist chart engines Sankey flow, Treemap (squarified), Choropleth (GIS), Calendar Heatmap, Word Cloud and Venn / Euler. For each engine you get what it draws, a working example you can paste straight into the Studio, the full syntax, the options it accepts, what flowss checks or works out for you, its limits, and the messages you may see. It also explains where each chart is drawn (inside flowss or in your browser), how big a chart can get, how charts export, and which engine to pick. Charts that belong to a business or operations workflow, such as funnels, waterfalls, Pareto charts and control charts, are documented on their own reference pages and are listed in Charts documented on other pages.

At a glance

Every engine on this page is available on every plan, including Free, and the engines themselves are not metered. (Free does limit how many figures you keep, to three per studio.) Only the optional AI features, such as flowss Studio Agent: pick the best chart in Data → Chart, Explain & fix, Fix with the flowss Studio Agent and automatic repair, need AI: there is no AI on Free, your own AI key works from Starter, and hosted AI points start on Plus. See Plans and what they include.

Working with chart engines

Opening an engine

1. In the Studio, click Engines & templates at the top of the tool rail (its tooltip also names the current engine), or the engine name at the top of the code sheet. The library opens with two tabs, Engines and Templates. (Pressing / opens the same library on the Templates tab with the template search focused.)
2. On the Engines tab, search by name or by what you are making, for example "scatter", "histogram", "sankey", "treemap", "map", "heatmap" or "word cloud", or by the tool you are replacing, such as "Excel charts", "GraphPad Prism", "Tableau" or "Datawrapper".
3. Click the engine. Its starter (its first template) replaces the current figure's source. The switch is one undo step, so ⌘Z (Ctrl+Z on Windows and Linux) brings your source back. Start a New figure (⌥⌘N) first if you want to keep both.

You can also click Open in Studio on any engine's reference page, which opens /studio?engine= followed by the engine id, or filter the template gallery by engine, for example /templates?engine=plot. Every reference page shows the engine's template count, its source language, the studio it opens in and its engine id, a Syntax at a glance sample where one exists, and up to six sample templates with a link to all of them.

Three ways to get a chart from data

Data → Chart is documented in full, including its recommendation rules, chart builder and limits, in Data, timeline, TikZ and the diagram tools.

Where each chart is drawn

None of the engines on this page uses an outside render service, so none of them is subject to the daily outside-service render allowance.

Tip: If a chart has to be drawn on a server, for example in a CI pipeline or for an automatically generated report, write it in Plot (for XY charts) or one of the native chart engines rather than in Vega-Lite or Plotly.

Size budgets

Before it draws, the Studio checks the size of the source against a budget for the engine, so a pasted giant cannot freeze the page. A source over budget is not drawn.

Bytes are counted in UTF-8, so accented letters and symbols such as ± and ° count as two or more. The message reads "Source is … — past the … safe-rendering budget for … Rendering this in the browser would freeze the page." followed by three options: split the figure into smaller figures, render it on the server with the render API, or trim the source. The server option applies only to engines drawn inside flowss. Most native engines also have their own, smaller caps on what they draw (listed with each engine below); those are reached first, and the figure says which cap was hit.

Checks and readiness

Every engine on this page has a checker. Its findings appear in up to three places:

• Inside the figure. Plot prints a notes strip under the chart, Sankey flow and Treemap print a headline and their most important findings under the drawing, and Venn / Euler prints notices for anything it could not draw. Exports carry these lines.
• The readiness chip in the Studio's status panel. A chart whose checks fail cannot read as Sound. Open the chip's detail to see what is Blocking or Holding it at a level, and use Fix beside a finding to hand just that one to the flowss Studio Agent (AI needs to be on your plan).
• Diagram Intelligence (⇧⌘X) adds metrics and insights for Sankey flow: nodes, flows, sources, sinks, throughput, whether flow is conserved, leaking or gaining nodes, cycles and the largest flow. See Flow-systems analysis in Studio.

A checker that could not read part of a document never presents its result as complete. It names the line it could not read, and verdicts such as "clean" are withheld until the whole document has been read.

When a chart fails to draw at all, a Render error box appears over the canvas with the message. If the message names a line, Go to line followed by the number puts your cursor on it. Explain & fix sends the error and the source to AI and applies the fix, which you can undo.

Exporting charts

Charts export like any other figure from the Studio's File menu (the ⋮ button), under Export:

A few chart-specific details:

• Plotly PNG and PDF exports use Plotly's own image exporter, at least 800 × 480 pixels before scaling, so text and lines stay crisp.
• Vega-Lite and Vega are drawn as SVG, so every export path works and the SVG stays editable in a vector editor.
• The native engines produce SVG directly, which is why they also work in the render API, embeds and thumbnails.

To put a chart in a LaTeX paper, Export as TikZ / LaTeX (⌘L) converts any figure into a standalone LaTeX document with AI. It needs AI on your plan, and sources over 30 KB cannot be converted; see Data, timeline, TikZ and the diagram tools. Every export format is covered in Exporting your work.

Plot

Plot draws plain XY charts and plotted functions: scatter plots, line charts, grouped bars, histograms with a named bin rule, Tukey box plots, and formulas sampled adaptively so a fast-changing curve is not quietly drawn as a different, smoother one. It is drawn inside flowss, so it works offline, on the server and in embeds. It is the right engine for "here are some x values and some y values, draw them", and for any figure where you want to plot a formula rather than type in points.

title: Cooling curve
subtitle: Cabinet A, 24-hour soak test
x: Time (h)
y: Core temperature (°C)
grid: both
legend: right

scatter "Measured" colour: blue error: sd
  0, 21.4 ± 0.3
  4, 16.2 ± 0.4
  8, 12.9 ± 0.6
  12, 10.8 ± 0.5
  24, 8.4 ± 0.4

plot "Model" y = 8.2 + 13.2*exp(-x/7.4) for x in [0, 24] dashed: yes

note: The dashed curve is drawn from the design constants, not fitted to these points.

How a Plot document is written

A Plot document is line-based. Directives are written key: value. A series starts with a header line naming its kind, followed by its data rows; indenting the rows is conventional but not required. Blank lines are ignored. # and // start a comment anywhere outside quotes, with one exception so that hex colours can be written: a # that comes straight after a key: and is followed only by hex digits is a colour, as in colour: #0f172a, not a comment.

Directives

The unit in an axis title is only read as a unit when there is a space before the bracket. Time (h) has the unit "h"; sigma(tau) is a function of tau and keeps its bracket as part of the title.

Series kinds

The first series with data decides what kind of figure it is, because a figure cannot have a numeric and a categorical x axis at once:

A series that does not fit is not drawn and is named in the notes strip, for example "«Model» is a function series, and this figure has a categorical x axis (because the first series is a bar series). It was not drawn …". Move it to a figure of its own.

Writing data rows

Fields in a row can be separated by commas, semicolons, tabs or runs of spaces, which is what spreadsheets, CSV files and papers produce.

Any series can carry its data inline after a colon on the header line. Items are separated by commas, so inside one item use spaces:

bar "Revenue": North 12.4, South 9.1, East 7.8
scatter "A": 0 1.2, 1 1.9, 2 3.1
box "Before": 120 135 128 142 160 151 133

Series options

Options go on the series header, after the name and before any inline data, written key: value.

The option list stops at the first key: Plot does not recognise, so a misspelt option shows up as unreadable data rather than being silently ignored.

curve: monotone never overshoots, so the smoothed line cannot invent a peak the data does not contain.

Plotting a formula

plot y = sin(x) for x in [0, 2pi]
plot "Envelope" y = exp(-x/8) for x in [0, 30] samples: 400 dashed: yes
f 1/x for x in [-4, 4]
plot "Decay" y = 100*exp(-t/12) for t in [0, 60] colour: teal

The line starts with plot, function or f, then an optional quoted name, then the formula (the y = is optional), then for, the variable and the interval in square or round brackets, separated by a comma or semicolon. The variable can have any name, and the formula may use only that name. Names are not case-sensitive. The interval ends can themselves be expressions, such as 2pi, but may not mention the variable. On a function line the options come after the interval, and only three apply: samples:, colour: (or color:) and dashed: (or dash:). A function line takes no data rows; a row written under it is skipped with a note.

Note: log is base 10, as on a calculator. Use ln for the natural logarithm.

Plot samples the formula on an even grid (129 points unless you set samples:), then repeatedly splits any stretch that is still bending, so a narrow peak or a fast oscillation is resolved rather than smoothed away. It breaks the line wherever the function has no finite value and across asymptotes, so 1/x is not joined straight across zero. If refinement runs into the 4,000-sample limit, the notes strip says which curve stopped being refined and that "The curve you see is smoother than the function is."

Histograms and box plots

• Histograms choose their bin count by the Freedman–Diaconis rule. When the interquartile range is zero, Sturges' rule is used instead. The rule that ran is named in the notes strip, for example "«Batch 7»: 5 bins, chosen by Freedman–Diaconis." Set bins: to choose the count yourself. The last bin includes its upper edge, so every value is counted.
• Box plots are Tukey boxes: median, quartiles, whiskers and outliers. With fewer than about five values, a note suggests that a strip of points would be more honest. Two box series with the same name cannot both be drawn; the second is named and skipped.

Colours and palettes

Series take colours from the figure's palette in order, unless a series sets its own colour:. The categorical palettes you can name with palette: are okabe-ito (the default), tol-bright, tol-muted, tol-light, tol-vibrant, tol-high-contrast, tol-medium-contrast and greyscale-safe. Named colours have light and dark variants and follow the Studio theme.

What the figure tells you

Under the title, a summary line gives the series and points drawn, for example "1 series · 5 points". If any cap bit, the count is printed with ≥ and the word "capped". The notes strip under the chart lists, in order of importance, everything the chart could not draw or wants you to know, together with your own notes and caption. Plot raises these by itself:

• A cut bar baseline: "The bar baseline is …, not zero. A bar encodes quantity by its length, so a cut baseline makes small differences look large …". A histogram gets the equivalent warning: "The y axis does not reach zero on a figure whose bars encode quantity by length, which exaggerates every difference in it."
• Values a logarithmic axis cannot place: "The x axis is logarithmic and 2 values are zero or negative. A log axis has no position for them, so they are not on this figure at all …" (a similar note covers the y axis).
• An interval meaning Plot does not know: "… is not an interval this engine can name. The bars are drawn, but the figure cannot say what they mean — use sd, sem, ci, iqr or range."
• A colour Plot will not use: "… is not a colour this engine will use, so the palette chose instead."
• Intervals with no stated meaning: "«Observed» carries intervals with no stated meaning … add «error: sd» or «error: ci»."
• Rows that could not be read: "«scatter 1»: 1 row could not be read as data and was skipped."
• Lines outside any series: "«…» is not a directive and there is no series open for it to belong to, so it was skipped."
• Clipped titles and notes: the clipped part is printed in the strip so nothing you wrote is lost.

What Plot deliberately does not do: pie charts, dual y axes, stacked areas, 3D, interactivity, CSV loading, date parsing and regression fitting. Use the chart grammars below, or the Figure studio, for those.

Plot limits

Vega-Lite

Vega-Lite is a grammar of interactive graphics written as a JSON specification: you describe the data, a mark and how fields map to channels, and Vega-Lite works out the scales, axes and legends. It covers scatter, line, bar, area, heatmap, box plot, violin and density charts, layered and faceted charts, geographic maps and interactive selections. flowss uses Vega-Lite 6.

{
  "$schema": "https://vega.github.io/schema/vega-lite/v5.json",
  "title": "Monthly sign-ups",
  "data": {"values": [
    {"month": "Jan", "signups": 120},
    {"month": "Feb", "signups": 148},
    {"month": "Mar", "signups": 171}
  ]},
  "mark": "bar",
  "encoding": {
    "x": {"field": "month", "type": "ordinal", "sort": null},
    "y": {"field": "signups", "type": "quantitative"}
  }
}

How flowss draws Vega-Lite

• The chart is drawn in your browser as SVG.
• Any background colour in the specification is replaced with a transparent one, so the chart sits on the canvas. When you export, the background comes from the export settings (the theme's page colour, or none with Transparent background), not from the specification.
• The Studio's dark theme is applied to the chart when you are in dark mode.
• The Vega action menu (View Source, Open in Vega Editor and the export links) is not shown. Export from the Studio's export menu instead. Embedding options written inside the specification under usermeta.embedOptions are ignored, so a shared specification cannot turn that menu back on.
• The specification must be a JSON object. A specification that is only a string is refused with "A Vega or Vega-Lite spec must be a JSON object, not a string: a string spec is fetched as a URL."

Data

Put data inline with "data": {"values": [...]}. A "data": {"url": "..."} can load from flowss itself, from the Vega example datasets at https://cdn.jsdelivr.net/npm/vega-datasets@2/, and from raw GitHub files on raw.githubusercontent.com and gist.githubusercontent.com. Most other hosts are blocked by the page's security policy, so paste the values inline instead, or aggregate them first.

What flowss checks

flowss reads every encoding and follows the specification's transforms (filter, calculate, aggregate, bin, window, join-aggregate, density, quantile, regression, loess and others) to work out which fields exist at each point. It then reports:

Fields inferred without a type are not reported, because Vega-Lite infers them reliably. The count aggregate needs no field, and a nested field such as a.b is checked by its first part.

Templates and Data → Chart

The chart builder in Data → Chart writes Vega-Lite for Scatter, Line, Bar, Histogram, Box plot, Violin, Heatmap, Pie, Area and Density, with your rows embedded. Several recommended figures (lines over time, histograms, sorted bars, bars with error bars, part-of-whole arcs, heatmaps, box plots and scatters) are Vega-Lite too. The template gallery has well over a hundred Vega-Lite templates to start from.

Vega

Vega is the lower-level grammar under Vega-Lite. A Vega specification declares every data set, scale, axis, legend, signal and mark by name, which makes it the engine for bespoke visualisations: chord-style and radial layouts, sunbursts, force-directed networks, parallel coordinates and custom interactions. flowss uses Vega 6.

{
  "$schema": "https://vega.github.io/schema/vega/v5.json",
  "width": 300, "height": 160,
  "data": [{"name": "table", "values": [
    {"c": "A", "v": 28}, {"c": "B", "v": 55}, {"c": "C", "v": 43}
  ]}],
  "scales": [
    {"name": "x", "type": "band", "domain": {"data": "table", "field": "c"}, "range": "width", "padding": 0.1},
    {"name": "y", "type": "linear", "domain": {"data": "table", "field": "v"}, "range": "height", "nice": true}
  ],
  "axes": [{"orient": "bottom", "scale": "x"}, {"orient": "left", "scale": "y"}],
  "marks": [{"type": "rect", "from": {"data": "table"},
    "encode": {"enter": {
      "x": {"scale": "x", "field": "c"}, "width": {"scale": "x", "band": 1},
      "y": {"scale": "y", "field": "v"}, "y2": {"scale": "y", "value": 0},
      "fill": {"value": "#4c78a8"}}}}]
}

Vega is drawn exactly as Vega-Lite is: in your browser, as SVG, with a transparent background, the dark theme in dark mode, no action menu, and the same rules for data URLs.

What flowss checks

flowss first asks Vega's own parser whether it accepts the specification, and shows Vega's message when it does not (for example an undefined data set, an unknown scale on an axis, an unknown signal or a duplicate name). It then looks for the mistakes Vega accepts in silence:

Names are checked in the scope where they are written: a group mark's own data, scales and signals are visible to its children but not to its siblings.

Plotly

Plotly draws publication-quality scientific and statistical plots from a JSON figure: scatter and line, bar, box, violin, histogram, heatmap, contour, 3D surface and scatter, polar, ternary, waterfall, funnel, parallel coordinates, Sankey, tables, indicators and more. flowss uses Plotly 2.35. It is the engine most templates use for specialist scientific figures such as ROC curves, Kaplan–Meier curves, dose–response curves and forest plots.

{
  "data": [
    {"type": "scatter", "mode": "lines+markers", "name": "Treatment",
     "x": [0, 7, 14, 21, 28], "y": [100, 92, 81, 77, 70]},
    {"type": "scatter", "mode": "lines+markers", "name": "Control",
     "x": [0, 7, 14, 21, 28], "y": [100, 97, 95, 94, 92]}
  ],
  "layout": {"title": "Tumour volume (% of baseline)",
             "xaxis": {"title": "Day"}, "yaxis": {"title": "% of baseline"}}
}

What a Plotly document can look like

A document with no traces anywhere is refused with: "This Plotly document has no traces to draw. A figure reads {"data": [{"type": "scatter", "x": [...], "y": [...]}], "layout": {...}} — a bare array of traces, or a single trace, also works."

How flowss draws Plotly

• The chart is drawn in your browser, 900 pixels wide (or 95% of the canvas, if that is narrower) and 520 pixels tall, and resizes with the canvas.
• The paper and plot backgrounds are made transparent, and the font is set to the Studio's typeface and theme colour. Any layout.font you write (family, size or colour) is replaced by these settings. Margins default to 50, 30, 50 and 50 pixels (left, right, top, bottom); any margin your layout sets wins.
• Your config is applied (zoom, mode-bar buttons, locale), except that the Plotly logo and every link that would send the chart's data to Chart Studio are switched off.
• Plotly's own mode bar, with zoom, pan and its own image download, appears when you hover over the chart. For figures to share, use the Studio's export menu, which uses Plotly's exporter for PNG and PDF.

What flowss checks

Trace types that keep their data elsewhere (parallel coordinates and categories, scatter-plot matrices, Sankey, tables and indicators) are recognised and not reported as empty.

Sankey flow

Sankey flow draws multi-stage flow diagrams for budgets, energy balances, material flows, conversion funnels and traffic. Each ribbon's thickness equals its value, nodes are placed in columns by their depth in the flow, and each node shows its total throughput. Unlike a drawing tool, it adds the flows up: it tells you when a node does not pass on what it receives, and when the diagram as a whole loses or gains a quantity.

title "Household energy (kWh)"
node Grid
node Solar
node Home label "Home supply"
node Heating
node Appliances
node Export
flow Grid Home 6200
flow Solar Home 2100
flow Solar Export 900
flow Home Heating 4300
flow Home Appliances 4000

Syntax

• A node named only in a flow is created automatically (and noted).
• Two flow lines for the same pair add together, which is what two ribbons between one pair mean.
• Values may be decimals, and thousand separators are allowed (1,200). A negative value is refused, because a negative flow is a flow the other way: write it in the other direction.
• A unit after the number, such as 40.5 PJ, is not read; the value is used and a note says the unit was ignored, because a Sankey has one unit throughout.
• # and // start a comment outside quotes.

What flowss works out

A node with no inbound flow is a source and a node with no outbound flow is a sink; neither is an imbalance. Every node between them must pass on what it receives, to within 0.5% of what flows through it.

When everything balances, the headline reads, for example, "Conserved — 9200 in, 9200 out, across 6 nodes and 5 flows. The widths add up." The headline and up to four findings are printed under the diagram, with a count of any further findings.

Limits: 400 nodes, 2,000 flows and 6,000 lines. Anything past a cap is reported, never dropped in silence.

Tip: Mermaid also draws Sankey diagrams with its sankey-beta type (see Software and architecture engines). Use the Sankey flow engine when the numbers must balance, or when the figure has to be drawn on a server.

Treemap (squarified)

Treemap fills a rectangle with nested rectangles whose areas are proportional to their sizes, using the squarified layout so every rectangle stays close to square. Use it for disk usage, budgets, portfolios, market share and any part-of-whole hierarchy.

title "Disk usage (GB)"
node photos size 77
node videos size 200
node code size 42
node photos.2024 parent photos size 45
node photos.2023 parent photos size 32
node videos.movies parent videos size 80
node videos.shows parent videos size 120

Syntax

• Ids cannot contain spaces. Use dots, dashes or underscores. The label drawn in each rectangle is the last part of the id after a dot, so photos.2024 is labelled "2024".
• Write parent before size when a node has both. Sizes may be decimals.
• Each rectangle shows its label when it is wide and tall enough, and its size when it is taller still.
• Colours are chosen per node from its id and get slightly deeper with each level of nesting.
• Comments start with # or // at the start of a line.

What flowss checks

When there is at least one error or warning, a headline and up to three of them are printed under the treemap. A document with no nodes shows "Add nodes — node photos size 7700". Up to 2,000 nodes are read.

Choropleth (GIS)

Choropleth shades a built-in hex-tile map by value: one hexagon per country or US state, placed roughly where it sits on the map. Nothing is downloaded, so it works offline and prints cleanly. Use it for teaching maps and dashboard-style comparisons across countries or states.

title "GDP per capita, 2023 (US$)"
atlas world
scale "#fef3c7" "#dc2626"
legend "GDP per capita (US$)"
USA 76399
GBR 52432
DEU 53562
FRA 46315
JPN 33950
CHN 13136
BRA 10412
IND 2485
NGA 2184

Syntax

Values are mapped linearly between the lowest and the highest value. Every tile is printed with its code, and tiles with no value are drawn in a pale neutral colour. Once at least one value is read, a legend shows the ramp with the minimum and maximum values, under your legend caption. Hovering over a tile shows its full name and value. Comments start with # or // at the start of a line.

The atlases

Warning: The world atlas is a teaching-sized selection of 45 countries, not a complete map, and tiles are approximate positions rather than real borders. A code with no tile, such as UK or CHE, shades nothing.

What flowss checks

Warning: Numbers must not contain thousand separators: write 76399, not 76,399. The value is read up to the first comma, so USA 76,399 is drawn as 76 and nothing on the figure or in the checks says so.

Calendar Heatmap

Calendar Heatmap draws one year as a grid of days, seven rows by up to 53 week columns, in the style of a contributions graph. Each day is shaded by its value. Use it for activity, habits, incidents, sales or any daily series.

title "Commits in 2024"
year 2024
2024-01-05 3
2024-01-06 8
2024-02-14 5
2024-03-21 12
2024-06-30 1
2024-11-02 7

Syntax

• Weeks run Sunday to Saturday, top to bottom, with Mon, Wed and Fri labelled and each month labelled where it begins.
• Days with a value are shaded in four steps of green relative to the largest value, with a five-square Less to More legend whose first square is the empty colour. Days with no value, or a value of zero, are left empty.
• Comments start with # or // at the start of a line.

What flowss checks

Word Cloud

Word Cloud sizes terms by weight and packs them, largest first, into centred rows, colouring them in turn from a fixed eight-colour palette. Use it for themes, keywords, survey free text and brainstorm summaries.

title "What customers mentioned"
performance 40
reliability 32
"developer experience" 28
security 24
"release 2024" 12
docs 9
pricing 6

Syntax

• One term per line, optionally followed by a weight (a whole or decimal number of zero or more). A term with no weight gets a weight of 1.
• Terms can contain spaces without quotes (developer experience 28), but quote a term whose name ends in a number, so the number is not taken as the weight: "release 2024" 12. Double or single quotes both work.
• A negative number is not read as a weight: gamma -3 becomes the term "gamma -3" with the default weight of 1.
• title "…" sets the title. Comments start with # or // at the start of a line.
• Font sizes run from 16 to 56 pixels, scaled between the lowest weight (or 1, if that is lower) and the highest. With no weights at all, every term is drawn at 26 pixels.

What flowss checks

Venn / Euler

Venn / Euler draws one, two or three overlapping circles and places your labels in the right regions, stacking several labels in one region. Use it for overlapping responsibilities, audience segments and skill matrices. The circles are a fixed arrangement; they are not sized to the data.

title "Who owns what"
set A "Sales"
set B "Marketing"
set C "Product"
only A "Quotas"
only B "Campaigns"
only C "Roadmap"
intersect "A,B" "Demand gen"
intersect "B,C" "Launches"
intersect "A,C" "Pricing"
intersect "A,B,C" "OKRs"

Syntax

Set ids are usually single letters, but any short token works, quoted or not. A region line may come before the set it names. Comments start with # or // at the start of a line.

What flowss checks

If no set can be read at all, the figure says so instead of drawing an empty circle, for example "No set in this source could be read, so there is nothing to overlap — line 1 was not read: …".

Charts documented on other pages

These engines draw charts too, but they belong to a workflow documented elsewhere:

Choosing a chart engine

Tips

• Start from a template. Open the engine's reference page, or the template gallery filtered by engine, and replace the numbers with yours. Templates are worked examples that pass their own checks.
• State what your error bars mean. In Plot, add error: sd, error: sem or error: ci to any series with intervals. A reader cannot tell a standard deviation from a confidence interval by looking.
• Keep bars on a zero baseline. Setting y limits on a bar chart so the axis does not reach zero is allowed, but Plot will say in the figure that it overstates every comparison.
• Use ln for natural logs in Plot. log is base 10.
• Quote awkward labels. In Sankey flow quote ids with spaces; in Word Cloud quote terms that end in a number.
• Check a Sankey before you share it. If the headline says a node does not balance, add a ribbon for the loss rather than adjusting a number to make it disappear.
• Prefer native engines for automated output. Plot and the other native chart engines draw in embeds, thumbnails and the render API; Vega-Lite, Vega and Plotly need a browser.
• Big data belongs elsewhere. The chart grammars are budgeted at 80,000 bytes of source. Aggregate first, or take the data to the Figure studio.

Limits and known constraints

• Vega-Lite, Vega and Plotly cannot be drawn by the render API, which refuses them with needs-browser. Use the Studio or an embed.
• Vega and Vega-Lite data URLs are meant for flowss itself, the Vega example datasets on jsDelivr, and raw GitHub and Gist files. Most other hosts are blocked.
• Plot has no pie charts, dual axes, stacked areas, 3D, interactivity, CSV loading, date axes or regression fitting, by design. It draws at most 8 series, 2,000 points per series and 8,000 points in total.
• Choropleth covers 45 countries on its world atlas and the 50 US states, as approximate hex tiles rather than true borders. Codes outside the atlas shade nothing, and numbers cannot carry thousand separators.
• Venn / Euler draws at most three sets, in a fixed arrangement that is not proportional to the data.
• Calendar Heatmap draws one year per figure and uses a fixed green colour scale.
• Treemap ids cannot contain spaces, and the colour of each rectangle is chosen automatically.
• Sankey flow reads up to 400 nodes and 2,000 flows; Treemap reads up to 2,000 nodes.
• Size budgets apply before drawing: 300,000 bytes for Plot, 80,000 for Vega-Lite, Vega and Plotly, 150,000 for Choropleth, Calendar Heatmap and Word Cloud, and 60,000 for Sankey flow, Treemap and Venn / Euler.

Troubleshooting

Related pages

• Choosing an engine
• The engine catalogue
• Data, timeline, TikZ and the diagram tools
• Figure and Figure analysis and statistics
• Flow-systems analysis in Studio
• Business, strategy and planning engines
• Process, operations and quality engines
• Software and architecture engines
• Maths, science and academic engines
• Exporting your work
• Embedding and the render API
• The template gallery