Mavera DocsReading results like a researcher
What the badges, numbers, and confidence scores actually mean — and what to ignore.
Read a result like a researcher
Inspect the decision, confidence, disagreement, and full paper trail.
Choose Studies from the sidebar.
Good afternoon
Choose a surface from the sidebar to begin.
- 1Open StudiesChoose Studies from the sidebar.
- 2Open a completed studyChoose the included example result.
- 3Read the decision firstOpen the winner and practical next move.
- 4Read confidenceOpen the plain-language confidence label.
- 5Inspect disagreementFind polarization and suspicious unanimity.
- 6Open the paper trailInspect methods, sample, sources, and run lineage.
The three-second read
Every result carries the same three elements. Read them in this order:
- The headline — the finding in one sentence. Written from the data, not vibes.
- The grounding badge — how real the underlying audience is. This calibrates everything else.
- The numbers — computed by code from recorded synthetic responses. Never model-written.
Grounding badges, decoded
| Badge | Built from | Trust it for |
|---|---|---|
grounded | Your real data (interviews, reviews, CRM) | Directions AND specifics — quotes echo real customers |
partially-grounded | A mix of your data + category knowledge | Directions confidently, specifics cautiously |
category-modeled | Documented category knowledge | Themes, rankings, relative comparisons |
estimate | A labeled model estimate | A starting hypothesis, nothing more |
What synthetic panels are designed for
Relative questions: which message wins, which price triggers resistance, which objection surfaces first, which feature ranks where. If you're comparing A to B, you're using the tool as designed.
Not: absolute market sizing from panel answers ("34% of Americans would buy" — no). For sizing, use a Program, which pulls real cited web data instead.
Confidence and realism scores
Studies return a confidence score (how internally consistent the responses were) and a realism score (how human the response patterns look). Below ~0.5 on either: treat as a pilot, sharpen the audience or questions, re-run.
Unanimous results deserve suspicion
If 100% of a panel says the same thing, the question may have had an obvious answer or the modeled audience may be too narrow to disagree. Variation in the open-text reasons is useful directional evidence, not proof of market prevalence.
The paper trail
Click any result in My Research → the full transcript shows every step: what was asked, what came back, how each number was computed. Every external claim in Programs/Audits links its source. If a number can't show you its work, Mavera doesn't show you the number.