Inspect the dashboard
- Open Settings → Usage and choose Refresh when you need updated local records.
- Select 7 days, 30 days, or All time.
- Compare the mode, daily trend, harness, credential/route, and model/provider breakdowns.
- Expand Task, turn, and model accounting for detailed observations and coverage.
- Choose Export CSV when you want to analyze the displayed usage outside Rudr.
Use the same period and route when comparing values. Chat, Code, Voice, and speech can use different connections even when they share a provider name.
Understand the cost labels
| Label or state | What it means |
|---|---|
| Reported cost | A cost observation supplied through the supported runtime path |
| Estimated cost | A calculation using available usage and pricing information |
| Subscription equivalent | An API-like equivalent for comparison, not an additional subscription invoice |
| Unpriced | Usage exists without enough price information for a cost |
| Covered by native aggregate | Aggregate cost exists, but model-level allocation is unavailable |
| Incomplete or partial | At least one contribution lacks full coverage; totals can understate work |
| Pending or provisional | Accounting has not reached a final persisted observation |
| Legacy | Older records do not have the same verification coverage |
The partial marker ≥ indicates that the reported total is not complete. Missing cost is not zero cost. Do not compare a subscription equivalent to an API invoice as if they were the same charge.
Investigate inconsistencies
Expand Diagnostics for pending records, persistence problems, and harness details. Check whether the affected run is still active, whether its route changed, and whether historical records are incomplete. Exporting does not repair missing observations.
Use remaining provider usage for account windows and cost warnings for threshold behavior.