# Uptime percentage and SLA

Uptime percentage answers "what share of the time was this monitor up" over a window you choose. An **SLA target**
is the percentage you promise (for example 99.9%), and HostTracker shows whether you met it and how much downtime
you still had room for.

## How uptime is calculated

For a monitor and a time window:

```
uptime % = time up / (time measured - downtime inside maintenance windows)
```

- **Time measured** is the time the monitor was actually being checked in the window. Time before the monitor
  existed and time it was paused are not counted at all.
- **Time up** runs from each confirmed recovery to the next confirmed outage; **downtime** runs from a confirmed
  down transition to the confirmed recovery - the length of each [incident](/incidents/what-is-an-incident/).
- **Maintenance**: downtime inside a [maintenance window](/maintenance/what-it-suppresses/) that suppresses
  **Stats** is left out of the calculation entirely - it counts as neither up nor down. Up time inside such a window
  still counts as up. A window that suppresses only alerts does not change the numbers.

The API returns the parts separately, so you can see exactly how a figure was made:

| Member (`GET /monitor/result/summary`) | Meaning |
|---|---|
| `totalSec` | Time measured in the window. |
| `upSec`, `downSec` | Time up and time down. |
| `maintenance.upSec`, `maintenance.downSec` | The part of those that fell inside maintenance windows. |
| `downSpans` | How many down periods. |
| `uptimePercent` | The result. Use it as given rather than recomputing it. |
| `slaTarget`, `slaMet`, `errorBudgetSecRemaining` | The SLA view, when a target applies. |

Rows come back ordered by monitor id, and the endpoint has no `sort` parameter; to list monitors worst first, sort
the rows by `uptimePercent` yourself (rows without it had no data) - see
[Sort a summary, worst first](/reference/operator-reports-and-summaries/#sort-a-summary-worst-first).

## Why the same monitor shows different percentages

Each figure is its own calculation over its own window: today, the last 7 days, this month and this year are four
different numbers, and a single hour of downtime moves a daily figure far more than a yearly one. A 30-day window
has 43,200 minutes, so each minute of downtime costs about 0.0023%.

Very short outages can look longer than they felt: an outage lasts from the check that confirmed it to the check that
confirmed recovery, so the check interval and recheck time set the resolution. See
[Short outages recorded as long downtime](/troubleshooting/short-outages/).

## SLA targets

You can set a target in two places:

- **On a monitor** - the monitor's `slaTarget` field (set it through the API, for example
  `PATCH /monitor/{id}` with `{"slaTarget": 99.9}`). Uptime summaries for that monitor are measured against it.
- **On a status page** - the **SLA target** on the page's Settings tab (`settings.slaTarget`, Webmaster band and
  above). The public page then shows the uptime "against a 99.9% target", an **error budget** for the last 30 days
  (for example "1 h 12 min used of 43 min allowed"), and a 12-month compliance table visitors can expand and export
  as CSV, also published as `sla-export.json` - see [Embed and export](/status-pages/embeds-exports/).

The error budget is the downtime a target allows over a period: `(100% - target) x period`. At 99.9% that is about
43 minutes in 30 days; at 99.99%, about 4 minutes.

For a one-off comparison, `GET /monitor/result/summary?...&sla=99.95` measures against a target you pass for that
request only.

## Check the numbers yourself

1. `GET /monitor/result/summary?monitor=ID&from=START&to=END` for the totals.
2. `GET /monitor/incident?monitor=ID&from=START&to=END` for the outages behind `downSec` - their `durationSec`
   values add up to the downtime (an incident that started before the window counts only its part inside it).
3. `GET /maintenance?monitor=ID&from=START&to=END` for the windows behind `maintenance`.

MCP: `get_uptime_summary`, `list_incidents`, `list_maintenance`.

## Related

- [What a maintenance window suppresses](/maintenance/what-it-suppresses/)
- [Reading results and incidents](/incidents/reading-results/)
- [Embed and export a status page](/status-pages/embeds-exports/)
- [Short outages recorded as long downtime](/troubleshooting/short-outages/)