01 How to use this

• ▸Scan the grid first, not the KPIs. Your eye is a parallel anomaly detector — sweep the 24×18 matrix for the one off-color square (the amber dim_customer @ 03:00 cell). Green-everywhere means stand down before standup.
• ▸Read columns as runs, rows as tasks. A vertical amber streak = one bad run; a horizontal streak = one flaky task across many runs. The pattern tells you whether to re-run a DAG or quarantine a model.
• ▸The flame strip is your budget. The top-1% longest tasks are where SLA risk lives — if fct_orders creeps past its bar, the whole critical path slips. Watch it before it cascades.
• ▸Let the AI predictor pre-triage. The critical-path card flags the task most likely to breach tonight and proposes retry-vs-backfill, so you act on the projection instead of waiting for the red.
• ▸Drill before you escalate. Click any cell for the live log tail, retry boundary, and upstream lineage — confirm root cause in one click rather than tab-hopping to Airflow.
• ▸For your own org: if you can't find your worst run in under 5 seconds, your run history isn't dense enough — density is the feature, not the bug.

02 Watch the walkthrough