Summary
The Cerebro summary page shows how many components it watches, how many are at risk, how accurate its alerts are and how it learns.
Cerebro → Summary is the Cerebro dashboard: the failure risk of all watched components and how the model is doing.
Indicators
Indicator | What it shows |
|---|---|
Watched | How many components have Cerebro turned on, by type. |
High risk | How many are at high risk right now. |
Medium risk (watch) | How many are at the Watch level. |
Low risk | How many are normal. |
No prediction | How many don't have a prediction yet (only if there are any). |
Open alerts | Open predictive alerts. Click it to go to Predictive alerts. |
Precision | Of the alerts with an outcome, what percentage was useful. |
False positives | What percentage of the alerts were false alarms. It is highlighted in amber if it exceeds half. |
Precision and false positives are calculated from the alerts of the last few days based on the team's verdicts: Useful, I already knew, I acted and prevented it or an alert that ended in failure count as hits; those marked False alarm count as false alarms.
Charts
Risk by component: heat map of the last 48 hours, by hour. The more intense, the higher the 24-hour risk (or the higher the unusualness in anomaly mode).
Risk trend: the risk of all components over the last 7 days, and how many were at high risk in the last hour.
Highest risk
The list of the riskiest components, with their risk bar, their percentage and their level. At the top right, choose the Horizon: 1 h, 6 h or 24 h. It starts with your company's default horizon.
If there are no predictions with a percentage yet, the list sorts by unusualness versus history (anomaly mode).
Click a component to open its page, or View all devices for the full list.
Learning and model improvement
Learning: recorded failures and verdicts per week, over the last 12 weeks. They go in as labels in the next training.
Model improvement: how each model version improves over time. A hollow version is one that didn't replace the current one.
From here you can go to View all alerts and View all models.
Notices on the page
If a component type is still learning, you'll see a notice with how many failures are still needed for the percentage-based model. In the meantime, the risk comes from anomaly mode, “unusual behavior” is shown rather than a percentage, and it doesn't raise alerts on its own.