Skip to content

Plans and limits

Limit Free Scope
Projects per account 2 Account
Events per project per month 10000 Project
Sites per project 1 Project
Synthetic runs per account, month 10 Account
PR audits per project, month 5 Project

One accepted page view is one event. Two things follow from that word:

Accepted. A page view the SDK sampled out never reaches the collector and never counts. At the default sample rate of 0.25, an allowance of 10,000 events covers roughly 40,000 page views.

Page view. The interaction event that shares a page view’s measurementId is not charged separately. Soft navigations are page views — a single-page app that rewrites the URL five times during one visit produces five.

If your app pushes history state for something that is not a view change, that is a real cost. data-track-soft-navigations="false" measures only the initial document load; see Configuration.

Events. The collector starts returning 429 and stops accepting measurements for that project until the month rolls over. Nothing else is affected — the dashboard keeps working, and the data already collected stays.

The counter is exact, but the check reading it can be a minute behind, so the cutoff is approximate at the boundary. It errs toward accepting.

Projects, sites. The form refuses, at the moment you submit it. These are rare, transactional decisions and they are exact by construction.

PR audits. Charged per new commit, not per workflow run — re-running a job on the same SHA does not spend another one. Past the cap the API declines the report and the Action comments that it did, without failing your build.

Synthetic runs. The “run” button declines. This is the one limit that is deliberately not exact: each run spends a third-party quota, and the check is a count taken at the moment you press it, so two runs started at the same instant can both be allowed.

The sample rate is the thing to reach for when a site outgrows its allowance. Performance data is a population statistic — the p75 of 2,500 page views and the p75 of 25,000 are the same number for any site with real traffic. Halving the sample rate halves the cost and changes the answer by nothing you can see.

What sampling does cost you is the tail: a rare slow route with few views may stop producing enough measurements to have a meaningful p75 at all. If that is the thing you are trying to watch, sample it up rather than down.

Limits resolve most-specific-first — a per-project override, then a per-account default, then the plan. null at any level means inherit; zero is a real answer and switches the feature off rather than falling back to a default.