Group and Cohort Bands

Two boxed group headers sitting over one continuous row grid that shares a left label column MUST be rendered as a single table.

|        | Loss of Income | Other         |  ← two boxed bands
| (1)    | (2)   | (3)    | (4)   | (5)   |  ← full-width numbered tier
| 2019   | 1,204 | 887    | 310   | 95    |  ← one shared row spine

The decisive signal

A full-width numbered tier is proof of single-table semantics.

If (1)…(13) runs across the whole width, the page is asserting that all those columns describe one row. Splitting there loses the row identity that spans the groups — the fact that 2019's Loss-of-Income figures and its Other figures belong to the same year.

• The group bands MUST be rendered as spanning cells over their column ranges.
• The numbered tier MUST be kept as a header row, not dropped as decoration. See Barely-Existing Columns and Headers (§9.3.1) — dropping the numbered tier as noise also destroys the evidence that the table is one table.

Same-shape grids stacked under cohort bands, with no column headers anywhere:

FEMALE
| 25 | 1.02 | 1.10 |
MALE
| 25 | 1.08 | 1.15 |

One table per cohort. The band becomes that table's header row.

Fusing these into a single table loses the row headers: with the bands demoted to ordinary rows, nothing says which block is female and which is male.

The band is a header here precisely because nothing else is — it is the only thing naming what the rows below it are.

The two shapes look alike from a distance and resolve in opposite directions, so both checks are run deliberately:

• Does the left label column run continuously through both bands, unrepeated? → one table.
• Does each band restart the same labels under it, with no column headers printed anywhere? → one table per band.

Date bands and shading changes inside a single grid are a different question again — see Bands Inside the Grid (§9.2.1).