Published aggregates, and why they disagree
Each of these counts a different object: strikes against refineries only, strikes against refineries and terminals together, individual drone impacts, or strike events. None of them is a damaging-hit count, which is what the model needs, so the count below was built at the event level instead.
Damaging events, June to August 2026
One row per documented unit stop or capacity loss at a named plant. Attributions marked inferred are reconstructions: most reporting says a fire near a unit, not a unit destroyed.
Events counted in the strike rate but not as damaging
A log that shows only what it accepted cannot be audited for selection, so the exclusions are listed with their reason.