A matrix locates the same content by several attributes at once

What it is

A matrix structure seats the same object at the intersection of several attribute axes: topic × audience, product × region, course × difficulty. Location is “I know its values on these dimensions,” not “I know its parent.” Opening the cell shows the object once; it is not copied for each parent. It shares the “many dimensions” intuition with faceted classification, but here the claim is about how content is addressed as a structure type: the matrix treats the cell as a place, not a filter as a temporary view.

A matrix is not two trees side by side. Side-by-side is still two single-parent hierarchies. A matrix is one arrival, several-axis coordinates.

Why it happens

People often hold two incomplete clues at once (“expense policy for new hires”). A single tree asks them to throw one clue away to pick a parent. A matrix treats both as coordinates; the cell is narrower and more stable than either axis alone. The object remains one record, attached to coordinates rather than duplicated, so updates do not fork.

Those coordinates must already live on the object. Missing a dimension, the cell cannot form, and the matrix degrades into a broken tree: people conclude that combination does not exist. That precondition of the type is why metadata completeness is treated on its own later in the group.

Studying it

Use two-clue tasks to contrast a matrix with a single tree, and show whether simultaneous location is actually used.

• Paradigms: scripts that supply two attributes; one group must walk a tree (pick one dimension first), the other walks a matrix cell; then test whether the matrix still degrades gracefully when only one dimension is given.
• Independent variables: number of axes, whether the cell is a precomputed page or a live filter, whether the two axes are truly orthogonal.
• Dependent variables: steps and success under two clues, whether people understand the same object is reachable from both sides, whether single-dimension tasks are slowed by the matrix.
• Methodological note: if the cell is actually a filtered result page, the study is measuring faceted retrieval, not matrix structure. The operational definition of a matrix is that the cell has a stable address (bookmarkable, shareable), not merely a current filter state.

Where it stops holding

When only one dimension is meaningful, a matrix is a spare empty axis. If axes entail each other, the cell adds no information and the tree has merely been drawn as a table. Hypertext “see also” is not a second axis and cannot pad out a matrix. On a small collection a matrix mints many empty cells, which are read as “this kind does not exist,” a harm larger than the gain from cross-location.

Applying it

• Confirm that people will arrive with at least two attributes, then build shareable cells for those dimensions rather than offering only filters.
• Each object has exactly one legal coordinate per axis; cell pages show the same record.
• When only one clue is held, the matrix must still be browsable along that axis. Do not require both values before starting.
• Verify by reaching an object with two attributes, bookmarking the address, clearing filters, and opening the bookmark. If the bookmark is only “search with parameters” and the object’s coordinates are not stable, it is not yet a matrix structure.

Related

• Within the group: G1.09.1 Hierarchy organizes content by parent–child links and is the usual default · G1.09.3 Hypertext organizes by associative links, not a single hierarchy · G1.09.4 Sequential structure forces a fixed order, as in tutorials and wizards · G1.09.5 Most systems keep hierarchy as the backbone and add hypertext laterals · G1.09.6 Structure type and organization scheme are independent dimensions · G1.09.7 A matrix demands complete metadata; missing fields block location
• Adjacent: G1.06 Faceted classification · G1.05 Metadata · G3.10 Faceted navigation
• Search terms: matrix structure · polyhierarchy · cross-classification