templates/focus-group.md
Lens: Focus Group (group format)
When: group discussion (not 1-on-1). A fundamentally different coding unit than in all other lenses. Coding unit: NOT the utterance alone, but utterance + speaker's position within the group (who agreed, who challenged, who stayed silent). Group dynamics are part of the data, not noise. Precondition before coding: the transcript MUST distinguish speakers by name/label (Speaker 1, Speaker 2…). If the transcript isn't diarized — this is an S1 blocker, not something you can code by eye; return it for re-tagging. Capture discipline: record separately (a) the individual opinion as first stated, and (b) how it changed under group influence. Conformity and social pressure distort the later part of the discussion more than the start.
Layer 1 — operational (facts + quote + line + speaker)
- G1 Group composition and roles — who took part (role/segment, not name, if anonymization is needed), whether a clear opinion leader emerged.
- G2 Topic and stimulus material — what was discussed, what was shown/asked as a stimulus.
- G3 Individual positions (per speaker) — each participant's first stated opinion on the key question, BEFORE the group influenced it.
- G4 Agreement/disagreement (reaction map) — for each key claim: who supported it (explicitly, in words), who challenged it, who stayed silent. Format:
Claim X — for: Speaker 1, 3; against: Speaker 2; silent: Speaker 4. - G5 Opinion shift during discussion — cases where a participant changed position after someone else's remark; "before" and "after" quote.
- G6 Dominance / suppression — who spoke more/less than others; were there interruptions, cut-off thoughts.
- G7 Spontaneous consensus — topics where the group aligned WITHOUT the moderator, naturally.
Layer 2 — analytical (conclusion + quote; CAPS) — to council
- A1 Real consensus vs conformity — where agreement reflects a genuine shared view, and where it's a group-pressure effect (check: was there a divergence in G3 before discussion that disappeared by G4). (unstable — latent)
- A2 Minority voice — a position stated by one participant and not backed by the rest, but worth attention (don't confuse with noise).
- A3 Opinion leader's influence — how much the discussion's course was shaped by one dominant voice (G6), whether this skews the picture of the group as a whole.
- A4 Fault line — which trait (role/segment/tenure) splits the group's opinion — this is often more valuable than consensus.
- A5 Method limitation — what from this group CANNOT be generalized to the whole audience (n=1 group, composition effect, moderator effect).
- A6 Key insight — 1 sentence: on topic [X] the group shows [consensus/split/conformity], the key divergence is [Y], generalize with caution on — [Z].
Forcing questions (for the MODERATOR — keep voices balanced)
Who thinks differently? [To a silent participant] what do you think? Before [Speaker] said this — what was your opinion? Is there anyone here who strongly disagrees?
Difference from S3 (reliability council)
Here, group consensus is DATA (the group's own position), not an artifact of several independent AI coding runs. Don't confuse the two: the S3 council is still needed separately — for unstable Layer 2 cells (A1, A3), run independent interpretation passes as usual.