Use this guide when dashboard values appear stale, missing, inconsistent, or difficult to interpret.
Diagnostic Flow
Common Issues
Charts show no data
Possible causes- Time window has insufficient recent activity.
- The selected dimension has no current records.
- Data pipeline delay between event ingestion and chart aggregation.
- Switch to LAST 30 DAYS and re-check.
- Validate the corresponding project/model has active traffic.
- Re-open after a short interval and confirm if data appears.
Delta looks incorrect
Possible causes- Comparing a low-volume period against a high-volume baseline.
- A sudden short spike or outage skews percentage changes.
- Compare the same metric across 7-day and 30-day windows.
- Pair delta interpretation with absolute totals, not percentages alone.
Latency up while API calls are flat
Likely interpretation- Efficiency regression, model change impact, or infrastructure contention.
- Inspect affected projects in Observability.
- Check recent model/deployment changes.
- Validate cluster and endpoint health.
Token output spikes unexpectedly
Possible causes- Prompt changes increasing completion length.
- Upstream workload shift toward long-form tasks.
- Compare input/output trend behavior by model.
- Review prompt templates and response limits.
- Re-check model routing for the affected project.
Escalation Checklist
Escalate with this context to reduce turnaround time:- Metric(s) affected
- Selected time window
- First observed timestamp
- Projects/models impacted
- Recent deployment or routing changes
Preventive Practices
Run a short daily dashboard review at a fixed time.
Maintain weekly baseline snapshots for key metrics.
Track post-change metric movement after every significant rollout.