[PerryLink]

Problem

A choices list with group is a paper question: src/client/submission-registry.ts:74-87 registers it as a radio member and grades it from one answer (the GenuiChoices.answer type is number | string). The renderer, however, picks its control type from the authored mode (src/client/blocks/choices.tsx):

type={mode === 'single' ? 'radio' : 'checkbox'}

So mode: 'multi' + group renders checkbox rows whose second pick can never be recorded — the shared answer stores a single label — plus a "pick any that apply" hint and data-choice-mode="multi".

Repro (jsdom component test):

x renders single-select controls even when the spec carries mode: multi
  -> expected [ 'checkbox', 'checkbox' ] to deeply equal [ 'radio', 'radio' ]

Fix

Derive the presented mode from the same rule the registry uses: a grouped list is single-select. single = group !== undefined || mode === 'single' now drives the input type/name, the container role, the hint, the submitted mode and data-choice-mode. Lists without a group keep checkboxes and local multi-select unchanged.

I did not make this a validation error: rejecting the spec sends the whole fence back to a code block, while the node degrades safely to the single-select it is actually graded as. Supporting multi-select paper questions would be a feature change to the registry/grading contract, not a fix. (The only observable DOM change is that data-choice-mode now reflects the presented mode rather than the authored one; nothing in the repo consumes it.)

Tests

tests/genui-choices.spec.tsx:

1. grouped mode: 'multi' renders radio/radio, role="radiogroup", and data-choice-mode="single" (red on base);
2. the same holds in an export/panel context without block state (no answers provider) — the rule keys on group, not on the live registry (red on base);
3. a local multi-select list without group still renders checkboxes (regression guard).

npx vitest run tests/genui-choices.spec.tsx -> 16 passed; full suite 2109 passed / 152 files; npx tsc -p tsconfig.json --noEmit clean.