turning-engineering-analytics-into-insights

Query the GitHub warehouse tables or the engineering_analytics_* views with HogQL, test the query with the execute-sql MCP tool, then save it using insight-create with a DataVisualizationNode query. You can then bundle insights into dashboards with dashboard-create.

Call the engineering-analytics-sources MCP tool, which returns each connected GitHub source with its id, repo, and prefix. Table names are the prefix plus github_pull_requests, github_workflow_runs, github_workflow_jobs, github_reviews, and github_team_members.

Alerts require a trends insight, so a SQL insight cannot be alerted on directly. Instead, create a scheduled subscription for regular delivery, or use a threshold check inside a prompt-kind AI subscription.

No. The runner-tier price ladder lives in product code and drifts over time. Query the engineering_analytics_job_costs view instead, which renders that cost model into SQL and is parity-tested against the product's own tools.

The PR snapshot keeps only the current head SHA, so a SHA join silently drops every push except the latest. The run's pull_requests association gives the PR number, which correctly attributes CI across all pushes; head SHA is only valid for a PR's current CI status.

The view reads the Logs product, so it is bounded by short log retention and suits only recent windows. Fingerprinting is pytest-only and the data is failure-only, so report absolute counts rather than rates.