Here is a problem anyone who has drawn a chart has caused at least once. You plot a mean with an error bar. The error bar is computed from four observations. It looks exactly like an error bar computed from four hundred, only wider — and “wider” reads to a viewer as more variable, not as we barely know.
jk-plots are an attempt to make the second thing visible.
The paper was posted to arXiv on 22 October 2026 (arXiv:2610.31050) by Nikolaus Piccolotto, Daniel Pahr, Laura Lotteraner, Fatih Öztank and Markus Bögl. Its Comments field reads only “5 pages”, with no venue and no journal reference, so this is an unreviewed preprint and a short one.
The specific thing being communicated
The paper states the target precisely:
An important task of uncertainty visualizations is to communicate how the number of available observations influences the precision of estimates, such as a Normal distribution’s mean.
That is narrower than “show uncertainty”, and the narrowness is the contribution. The quantity being drawn is not the spread of the data. It is how much the sample size limits what you can say about the mean. Those are different, they move independently, and standard idioms conflate them.
What the plot does
Two ideas combined:
- Plausible reference models, drawn in a way that emulates popular uncertainty visualisation idioms — so the vocabulary is already familiar to a reader who has seen gradient intervals, violin plots or hypothetical outcome plots.
- The observations themselves, shown alongside.
And the approach is adaptive: the representation changes with the data, producing “salient representations while highlighting the uncertainty introduced by the sample size”. So the chart does not look the same at n = 5 and n = 500 beyond a change of width. The form itself carries the information.
The inspiration is named and is a nice detail: educational material from Bayesian statistics. These are teaching illustrations — the pictures used to explain to a student why a posterior is wide — promoted into a general-purpose chart type. That is a sensible place to look. Teaching diagrams are optimised for exactly the thing production charts fail at, which is making the shape of an inference legible to someone who does not already know the answer.
What exists
- A described visualisation pipeline for constructing jk-plots.
- An implementation as an R package.
- Demonstrations on synthetic datasets of varying size.
- Domain-expert interviews discussing usage scenarios.
The ordering there is the honest summary of the evidence. Synthetic data of varying size is the right way to show that a method responds to n, because you control n. It is not evidence that readers interpret the result correctly — that would need a graphical-perception study with human participants and measured accuracy, which this five-page paper does not claim to have done. The expert interviews discuss where you might use it, not whether it works.
The arXiv record lists only the abstract and PDF. No repository or CRAN link is advertised, so the R package is described rather than linked from the listing; we checked and there is nothing further to verify from there.
Why creative coders should care
Two reasons, neither of them statistical.
The first is that adaptive encoding is an underused idea generally. Most chart libraries give you a mark whose form is fixed and whose parameters vary. A mark whose form responds to a property of the data — here, how much data there is — is a design move that generalises well beyond uncertainty, and it is the kind of thing easier to build in p5.js or D3 than to retrofit into a charting package.
The second is that this is a rare example of a visualisation paper being clear about which of two confusable quantities it is drawing. If you are building a generative or data-driven piece, “is this mark showing variation in the world or my ignorance about the world?” is a question worth being able to answer about your own work, and most pieces cannot.
One practical caution: a chart that looks different at different sample sizes is harder to use in a small-multiples grid or an animation across time, because the reader has to re-learn the mark. The paper’s framing as adaptive is an advantage for a single considered figure and a cost for a dashboard.
Sources
- Adaptive Uncertainty Visualization With jk-Plots — arXiv:2610.31050
- arXiv PDF
- arXiv listing for cs.HC
- The Comprehensive R Archive Network