Gantt and Sankey Charts

The time axis is the value axis. Each bar's two edges are read against the printed ticks and written as the periods the axis prints: months as month tokens (Jan), quarters under year headers as quarter tokens with their year (Q1 2025). When the page prints exact dates for each task, those printed dates are the start and end; when it prints them as a table beside the bars, that printed table is the rendering and the bars add nothing to it (see Chart With a Printed Data Grid (§9.1.2)).

Nothing on the page prints the words start or end, and they MUST NOT be invented as labels of a mark; the pair lives in the row's two period cells.

How a gantt row is laid out. A gantt chart MUST be one table with one row per task bar or milestone, in the chart's top-to-bottom order: the printed label, the start period, the end period. Its header row MUST be Start | End over the two period columns, with the first cell holding the printed label column's title or left empty. These header cells are protocol vocabulary, like the markers [blank] or [logo: …]: the page does not print them, and they are the one deliberate exception to never invent, so that the columns can be read without guessing. They are written exactly as given.

A bar edge on a period boundary. A bar that fills January through March has its right edge on the April tick. Its end MUST be the last period the bar visibly fills (Mar), and one reading MUST be used throughout a document (see Choices Made Once Per Document (§4.2)). A checker accepts a reading one boundary off.

A flow's value is the number it carries. Where only node totals are printed, a node with a single inflow or a single outflow gives that one flow its value, and the flow takes the same row shape. A node that both splits and merges has no single flow its total belongs to: its total gives no flow row, and its printed total is kept as text. A sankey that prints no values has no readable flows: ribbon widths MUST NOT be estimated.

Node names in the source and target cells are copied without any printed node total. A ribbon is traced from end to end before its row is written — values often print at the ribbon's start, far from the target node, and a thin ribbon can pass under several others. A ribbon that cannot be traced unambiguously gets no row.

How a sankey flow is laid out. A sankey diagram MUST be one table with one row per flow, ordered by source from top to bottom and then by target, under the header row Source | Target | Value. Each printed node total that is not any one flow's value follows the table as text, as printed (Industry 165). These header cells are protocol vocabulary, like the markers [blank] or [logo: …]: the page does not print them, and they are the one deliberate exception to never invent, so that the columns can be read without guessing. They are written exactly as given.

A gantt span and a sankey flow are judged on meaning, not on a table shape: a checker accepts any output that conveys both endpoints of a span at the granularity the axis prints — a positional grid correctly aligned to the periods included — and accepts a span end one period boundary off. A flow is accepted wherever the output conveys its source, its target and its value together. A flow or span whose parts are scattered so that they can no longer be paired is not conveyed.