Sankey diagrams encode conserved flows as band width

What it is

A Sankey diagram represents states, processes, or accounts as nodes and transfers as directed bands whose transverse width encodes flow. It is most interpretable when one unit moves through a system with explicit boundaries: readers can trace origins and destinations and audit whether each node balances. Conservation does not mechanically require inflow to equal outflow at every node. Generation, consumption, loss, inventory change, sources, and sinks are legitimate when they appear as named, unit-bearing balance terms rather than unexplained gaps in width.

Why it happens

For a stated interval, a node account can read “opening stock + inflow + generation = closing stock + outflow + consumption or loss.” If stock is omitted, the chart represents transfer rates or interval totals and must state that boundary. With one linear width scale, incoming, outgoing, and explicit residual terms add geometrically, making the node an auditable ledger. Colour and path position help trace categories, but transverse width—not the band's total area—carries magnitude. A longer route must not imply more flow merely because it occupies more ink.

Studying it

Reconstruct a node-balance table from source records before testing the graphic. Check units, time windows, duplicate counting, split and merge rules, and unclassified residuals. Then compare Sankey and tabular views on source tracing, destination decomposition, loss localization, and balance auditing. Manipulate explicit loss, stock change, residual size, path length, and labels; measure accuracy, time, confidence, and area-based misreadings. Domain review of additivity and boundaries remains necessary: fluent band tracing does not establish a correct account.

Where it stops holding

Links that encode only association, similarity, or shared membership rather than additive transfer belong in a relation diagram. Probabilities and shares can use width only with consistent denominators and conditioning scopes. A user journey with exits, repeated visits, and returns across windows must distinguish exits, event counts, and entity flows rather than calling all of them “users.” Represent signed flow as a nonnegative band in its actual direction or as separate positive and negative accounts. Give missing destinations an explicit unknown node. Measurement imbalance is uncertainty or unexplained residual, not automatically genuine loss.

Applying it

• Define source, destination, unit, interval, and counted entity for every band, plus generation, consumption, loss, stock change, and unknown residuals for each node.
• Run automated node-balance tests before rendering and block discrepancies beyond a domain tolerance. Derive tolerance from measurement and decision risk, not one global threshold.
• Expose node totals and balance terms in labels, legends, downloads, and screen-reader descriptions so loss, unknown, and true zero remain distinct.
• Validate both path tracing and node-account auditing. Any width that cannot be recomputed from the underlying table returns to data modelling for repair.

Related

• Same group: U2.12.2 Node ordering and crossing count decide readability · U2.12.3 With huge magnitude gaps, small flows vanish · 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: U2.13.2 Layout algorithms shape node-link readings · U2.17.1 A mismatched chart forces readers to compute the answer mentally
• Search terms: Sankey diagram · flow conservation · node balance · stock and flow · residual flow