Currently these events are inferable from others in the log--most of the time. Explicit logging would provide more accurate data.

Since sampling of the banner history log occurs early on during the pageview, and it's likely that most of the time the banner close event occurs afterwards, pageviews that were included in the sample should be re-sent with the same log ID following such an event. Recall that the log ID is not persisted, however it remains available in memory for the duration of the pageview. By de-duping based on log ID, we'll get an accurate view of which pageviews had a banner close event.