No. It indicates lower model certainty, which may or may not correspond to an actual error.
Yes. Thresholds can be tuned to match different review strictness levels.
No. The heatmap is a review aid and does not modify subtitle content.