With huge magnitude gaps, small flows vanish

What it is

When Sankey flows span a wide dynamic range, a shared linear scale renders the tail as very thin bands. A flow may remain in the data and node account yet become hard to detect, trace, or select on the delivered display. This is a tension between small-flow visibility and proportional fidelity. “Vanish” does not imply a fixed flow ratio or pixel cutoff: contrast, path length, crossings, pixel density, zoom, task, and visual ability all affect performance.

Why it happens

One linear coefficient maps flow to width. Once large flows consume the available space, small ones shrink in true proportion. Raster rounding, antialiasing, and overlap can remove their boundaries before attention and motor limits make the surviving bands hard to trace or click. Log widths or a common minimum width reveal the tail but break additive width and geometric conservation. Dark strokes and highlighting can improve detection; they cannot make an exaggerated width continue to mean the true magnitude.

Studying it

At target size and device, manipulate flow ratios, number of thin bands, crossings, colour or pattern, path length, and interactive magnification. Test presence detection, end-to-end tracing, selection, magnitude comparison, and tail summation separately. Record omissions, mis-selections, time, confidence, and keyboard reachability. Compare individual bands, traceable aggregation, a detail view with a newly shared scale, and tables on realistic long-tailed data. Visibility does not establish accurate width reading, and one device cannot yield a universal threshold.

Where it stops holding

Volume-based aggregation may hide high-risk, rare, or disclosure-critical categories; retain named entries, filters, or a separate view when their semantics matter. “Other” is defensible only when membership, total, and drill-down remain available, and it must not mix directions or units. Sampling is weak for a complete conservation ledger unless the sampling rule and full total remain visible. A focused detail may rescale every visible band together and announce the changed scale; silently magnifying one local band in the overview is misleading.

Applying it

• Render the real distribution at delivered size and audit detection, selection, and tracing. Choose aggregation, splitting, or detail from task consequences rather than one ratio cutoff.
• For a tail aggregate, expose its total, member count, rule, and accessible detail, and verify node balances before and after aggregation.
• Give critical thin flows direct labels, outlines, keyboard focus, and selection highlighting. If width is illustrative rather than proportional, label and exclude it from magnitude comparison.
• Provide a flow-sorted table or bar detail and expose every flow's direction, value, unit, and aggregation state to screen readers so low salience does not mean missing data.

Related

• Same group: U2.12.1 Sankey diagrams encode conserved flows as band width · U2.12.2 Node ordering and crossing count decide readability · U2.12.4 Backflow and loops disqualify the Sankey · U2.12.5 Band width is a length encoding; one scale for the whole diagram
• Nearby: U2.11.2 Deep-node areas get too small to read; cap the depth · U2.16.1 Tables beat charts when exact values are needed
• Search terms: Sankey long tail · small-flow visibility · flow aggregation · dynamic range