Coverage and partial cycles

Vendors fail. Rate limits bite, an API changes, a SERP provider has a bad hour. A monitoring cycle that hits five of six engines has produced real data about five engines, and the honest thing to do is report five engines rather than either failing the cycle or quietly presenting a five-engine aggregate as a six-engine one.

How partial coverage surfaces

• The cycle completes and is scored on what it captured.
• Every affected figure carries its actual run count, which is lower — so the band widens rather than the number lying.
• The report's coverage line states which engines contributed to the period and which did not.

Tri-state AI Overview presence

For Google surfaces there is a third state that matters. Best-in-class vendor detection of AI Overviews runs at roughly 68% — meaning about a third of genuinely-present AI Overviews are returned as nothing. Recording that as “absent” corrupts a trendline permanently, so presence is modelled as present, absent or not_detected, and the third state is excluded from rates rather than counted as a miss.