The quiet candidate in a paired task

No - but not because the situation is fine

• Comparability is the decisive argument. Every other pair in the session met the same script. If Janet prompts B, B's sample is produced under different conditions from every other candidate's, and the score is no longer comparable. The recording goes to moderation; a deviation will be visible and will be queried.
• The script permits two fixed prompts and one clarifying repeat. Those exist for comprehension problems, not for redressing turn-taking.

Address the real worry Janet's instinct is that B has not produced 'enough language'. That is a concern about evidence sufficiency, not about fairness in the moment. It is handled at scoring, and if the scale has no insufficient-evidence rule, that is a gap in the instrument to fix later - not a licence to improvise now.

What she can do inside the script Use the fixed follow-up to address B directly and by name; a follow-up is the one legitimate slot where a silent candidate gets a guaranteed turn.

Score what is there, against the criteria

• The 20-second turn is real evidence: it is coherent, developed, and shows she can sustain an idea when she has the floor. Score Range and Accuracy from it, cautiously.
• Interactive competence is a criterion in this task. B's four minimal turns and failure to claim the floor are evidence about turn-taking - she does not interrupt, does not build on A's ideas, does not repair. That is scoreable, and it is genuinely hers; A cannot be blamed for B never saying 'can I finish that thought?'

But discount for the conditions A spoke 78% and interrupted twice. Some of B's silence is environmental. When evidence for a criterion is thin and the partner's behaviour contributed, the defensible move is to score at the level the available evidence supports and note the constraint - not to infer ability she never displayed, and not to penalise her for a floor she never got.

Check before finalising if the scale has a documented insufficient-evidence procedure; if it does not, record the constraint in the examiner report so moderation sees the reasoning.

Change the task shape Add a short individual long turn before the paired discussion - 60–90 seconds per candidate on the same prompt. Every candidate then produces a comparable monologue regardless of partner, and the paired phase tests interaction rather than carrying the whole score.

Make turn-taking explicit If interaction is a criterion, define it so examiners can score it: 'initiates, responds, develops partner's point, manages turn boundaries'. Vague criteria produce the current drift - some examiners reward fluency, others reward politeness.

Control the pairing variable

• Pair strong with strong and weak with weak where numbers allow, so dominance is less likely to be structural.
• Train examiners on dominance patterns and on what not to do; the script is the safeguard, and examiners need to have internalised why it exists.
• Monitor: compare mean scores of candidates who had a dominant partner with those who did not. If the gap is systematic, the task, not the candidate, is being measured.