Analytics

Clave reads traffic live from the edge. There is no script in the site's HTML and no per-site object at any provider.

That design has limits, and Clave states them rather than hide them. Each one is structural.

Four limits

• 7 days of history. A longer period is refused. A "30d" that quietly serves a week is worse than an error.
• No unique visitors and no referrers. Both are denied at the provider on this plan.
• Crawlers count as traffic. Telling a bot from a person is an enterprise feature.
• Sampling. Every count is scaled by its sample interval, and the answer says sampled: true. A raw count would under-report in silence.

What the answer carries

The answer carries experimental: true and a caveat written in plain language. The caveat is meant to be relayed to the owner word for word.

Read it only when you draw it

The quota for these reads is a platform budget, shared by every agent on Clave. Seven days of history costs seven queries.

So no page reads traffic it does not draw. The browser app reads analytics on the one tab that shows a chart, and puts a count of messages under each site on its home page instead.

An agent should do the same: read analytics when somebody asked about traffic.

Read Analytics.