Skip to content

Matrix questions by voice

Three sections contain matrix questions and the rest do not. Loading this page into all eleven section subagents would spend context on a problem most of them never meet, so it ships only to the three that do:

Why it goes wrong

A grid cannot be read aloud. Dumping nineteen rows into one turn gets nineteen answers of "fine" — the client stops listening around row four and starts producing the shape of an answer rather than an answer.

Asking one row per turn fails the other way: nineteen turns of "and your shoulders?" is intolerable on a phone call and the client hangs up.

The cluster rule

  • Walk the rows in thematic clusters of four or five.
  • State the response options once per cluster, then just name the items.
  • Let the client answer several in one reply. This is the single exception to the one-question-at-a-time rule in Core rules, and it applies to matrix rows only.

For each of these, just tell me if it's more than normal, less than normal, or about right: appetite, portion sizes, snacking between meals.

Handling the reply

A cluster reply arrives unordered and incomplete. "The first one's fine, and I definitely snack more" leaves portion sizes unanswered.

  • Map what you got, then ask only for the gaps. "Got it — and portion sizes?"
  • Never spread one answer across the cluster. "All fine" covers the cluster only if the client heard every item; if they interrupted, re-name the ones they missed.
  • A cluster is not a section. Leaving one row blank is a skipped field, recorded as such, not a reason to re-run the cluster.

Food frequency specifically

For the grid in section 7.2, go by food group and accept rough answers. "Most days", "hardly ever" and "only weekends" are all usable — normalise them to the declared scale and keep the client's phrasing alongside.