Every StatFacts card carries a confidence badge. It is not a quality score for the writing—it describes how the number was produced.

Meta-analysis

What it means: Multiple studies or datasets were aggregated; the range reflects pooled evidence.

Strengths: Broader than a single experiment; useful for planning bands.

Limits: Publication bias, different populations mixed together, still may not match your product.

Use for: Roadmap prioritization, rough sizing, stakeholder education.

A/B test

What it means: A controlled experiment—usually on live traffic—with a measured lift.

Strengths: Closest to how product teams actually learn; often includes real UX context.

Limits: One company, one season, one segment; may not replicate for you.

Use for: Justifying a test you are about to run, comparing magnitude across tactics.

Study

What it means: Published research (observational or experimental) that supports a directional effect.

Strengths: Peer-reviewed or primary data sources; good for non-product domains (sports, health).

Limits: Causality is not guaranteed; lab conditions differ from your environment.

Use for: Hypothesis generation, category pages, science-backed hooks.

Estimate

What it means: A directional industry rule-of-thumb—useful, but not tied to one cited experiment on the card.

Strengths: Fast orientation when no clean study exists; honest about uncertainty.

Limits: Wide variance in the wild; easiest label to over-trust.

Use for: Icebreakers in workshops, backlog grooming—not board-level forecasts without backup.

How we apply labels

Editors pick the best available evidence linked on the insight. When evidence is thin, we label it estimate rather than inflate confidence.

Practical weighting

| Your decision | Lean on | || | Ship a test this sprint | A/B test, then study | | Annual revenue model | Meta-analysis + your own baseline | | Twitter / newsletter hook | Any—cite the label honestly | | Investor memo | Meta-analysis or study + primary source PDF |

When in doubt, open the sources and upgrade or downgrade the label yourself.