Logs answer a question only when the query selects the right records. Fenwick Dispatch has success and failure rows, a lowercase status variant, an older failure, and one missing duration. Those are deliberate boundaries in this original training dataset. An account-wide total would miss the difference between a recent incident and an old event.
Use where to select records, project to choose columns, summarize to aggregate, and ordering to inspect a result. Equality has specific case behavior; a query that selects Failed does not automatically select failed. A global count over no rows also has a different shape from grouping by nonexistent values. The local parser supports this documented pipeline subset over fixed rows. It does not ingest telemetry, execute arbitrary Kusto, join tables, or reproduce a production workspace.
Match the time window, value semantics and aggregate to the incident question.