Runtime snapshots bring P50 cold starts into the 400 ms range.

Added a global edge-node topology view to the console.

Cross-region migration jitter in the scheduler is down by about 40%.

The observability dashboard supports custom sampling rates and strategies for capturing long-tail failures.

Fixed occasional missing root spans in distributed traces under high concurrency.

Improved metrics-query response times across large time windows.

Added multi-tenant network microsegmentation with tenant-scoped policy delivery.

Fixed an edge case that prevented image-pull timeouts from retrying in some regions.

The event bus is now in public beta, with backpressure and partitioned consumers.

Rewrote the documentation site and cut onboarding time roughly in half.

Improved contrast for several console charts in dark mode.