An availability heatmap compares the same measure across services and reporting windows. A missing observation is not a failed service, and a confirmed 0% result is not missing data.
Put this in widgets/metric_matrix.rs. For availability, each value is 100 * successful_probes / attempted_probes with a positive denominator and consistent probe definition. Omit unobserved cells deliberately and explain their absence. Supply translated service/window captions, title and summary, with a fixed 0–100 scale.
Validate cell bounds, unique row/column pairs and finite percentages before rendering. Unlike a calendar's missing-day behavior, sparse heatmap cells are not drawn. Retain exact successful/attempted counts because a percentage based on one probe is not equivalent evidence to one based on thousands.
Test no probes, failed probes, delayed telemetry and window boundaries. Use a colour-blind-readable scale plus an exact-value view. The gallery screenshot is a separate fixture.