Comment by runtimepanic

Observability that can produce causal explanations rather than just timelines. We have great tooling for logs/metrics/traces, but very little that helps engineers understand why a distributed system behaved the way it did. Automated causal graphs for incidents still feel like an open problem.