The public dashboard reports fleet activity directly from the rental provider's API. This post documents the methodology so the numbers can be read without guesswork.
Where the data comes from
Fleet telemetry is read from the provider's console API on request. There is no intermediate store and no manual adjustment step between the source and what renders on the page.
Three groups of figures are published:
- Active instances — machines currently rented, counted at read time.
- Machine inventory — total machines listed, whether or not they are currently rented.
- Earnings — provider-reported totals for the selected period.
Refresh behaviour
The dashboard endpoint is uncached and reads live on each request. A reload gives current figures rather than a snapshot from a scheduled job.
When the telemetry API is unreachable, the dashboard reports that state explicitly rather than displaying a stale figure.
What the labels mean
Utilization percentages describe the reporting window only. They are not a forecast, and they should not be read as a projection of future capacity.
"Active" counts instances in a running state at the moment of the read. An instance that started and stopped between two reads does not appear in either. For short workloads, instantaneous counts understate activity — period totals are the better measure.
Reading the earnings figures
Earnings are reported as the provider reports them, split by GPU and storage. They are gross figures. Operating costs, power, and hosting are not deducted, so they should not be read as margin.