Build user personas from evidence, not stereotypes

You are a UX researcher building personas for a product team. A persona is a compression of evidence, not a creative-writing exercise - every attribute must trace to something observed, and thin evidence must say so out loud.

<research>
[paste what exists: interview notes, support tickets, reviews, survey answers, sales-call notes, analytics observations. Label each chunk with a source ID, e.g. INT-3, TICKETS-Q2, REV-14]
</research>

<context>
THE PRODUCT + THE DECISION AHEAD: [what the team is building or deciding that these personas must inform]
SEGMENTS WE SUSPECT: [current hunches, so they can be confirmed or killed]
</context>

Before building, cluster the research by observed BEHAVIOUR and GOAL (never by demographic), and note where two suspected segments behave identically - identical behaviour means one persona, whatever marketing hopes.

Requirements:
1. Build 2-4 personas maximum. Each contains: goal (what they are trying to get done), context of use, key behaviours, pains and workarounds, decision triggers (what makes them buy/adopt/churn), and what they would never do.
2. Every attribute carries its evidence: the source ID and a short quote or observation. An attribute with no source does not ship.
3. No demographic decoration - age, photo, hobbies appear ONLY if the evidence shows they change behaviour toward the product.
4. Confidence rating per persona (strong / moderate / thin) based on how many independent sources support it, with the [NEEDED: ...] research that would firm up the thin ones.
5. One anti-persona: who the team might mistake for a user, and the evidence they are not.
6. For each persona, 2-3 "so-what" implications aimed at the stated decision - what this persona demands or forbids.
7. Name personas by their goal ("The Friday-night reorderer"), never by alliterated first names.

Output: cluster summary -> personas -> anti-persona -> confidence + research-gap table -> implications for the stated decision.

Grounding: the pasted research is the entire evidence base; where it conflicts, show both signals and say which is better sourced - never average them into fiction.

Copy the block above straight into Claude — anything in [BRACKETS] is yours to fill in.