The same numbers we use to run the network.

"Ad views" on Skrypte aren't a raw count of every ad tag that fired. Each impression is checked, at the moment it happens, against whether it was actually visible on the visitor's screen. That check looks at whether the ad element entered the viewport at all, and how long it stayed there before the visitor scrolled past or left the page. Impressions that never became visible, or that were on screen too briefly to plausibly have been seen, aren't counted as ad views.

Traffic quality is checked at the same time. Each impression carries a score used to flag activity that looks automated rather than human: patterns that don't match how a person actually browses a page. Impressions flagged this way are filtered out before they ever reach your dashboard or your revenue totals, so the traffic you're paid for is traffic the system believes was a real visitor looking at a real page.

This isn't a separate fraud report bolted onto the side of analytics. It happens before an event is ever aggregated into what you see. The viewability rate shown in your dashboard, the ad views total, and the revenue figure next to it are all downstream of the same filtered, per-impression data. There's no second, cleaner pipeline used only for the numbers publishers see, and no separate one used for billing. It's the same events, checked the same way, every time.