Quality
Sampling
Also called sampled report.
Sampling is computing a report from a subset of hits and scaling the result. It speeds up queries and costs accuracy.
How it is measured
The system picks a share of hits, say 10 percent, counts them, then multiplies. Small segments suffer most. The report should say it was sampled.
Tarsier stores hits in Cloudflare Analytics Engine, which can sample at very high write volumes. A small or mid-size site will see exact counts. A very busy one should check the volume before trusting small segments.
Worked example
A retailer filters a sampled report to mobile visitors from one city who bought a gift card. The base sample has 3 sessions, scaled to 30. The next refresh shows 60.
The merchant stops using that report and runs an unsampled one.
How it differs
Sampling estimates from some hits. Unsampled data uses all of them. Statistical significance tells you if a difference is real. Sampling widens the error bar.
Common errors
Filtering a sampled report down to tiny segments. Ignoring the sample badge. Comparing sampled and unsampled. Reading ratios from 3 sessions. Not knowing the rate.
In practice
Check whether your reports are sampled, and for small segments use an unsampled export.