How to calculate uptime when monitoring data is missing

A green percentage can hide a gap

Suppose a service is monitored over 60 minutes. You have 40 minutes classified as available, five minutes classified as unavailable and 15 minutes without usable observations. Reporting 40 divided by 45 gives 88.89% observed availability. Reporting 40 divided by 60 gives 66.67% confirmed available time. Neither number alone describes the complete situation.

The first answers how often the service was available during classified time. The second answers how much of the entire window is confirmed available. State the denominator. Monitoring coverage in this example is 45 divided by 60, or 75%. The missing quarter of the window should stay unknown until evidence supports another classification.

Define time states before calculating

Write down how your service treats available, unavailable, unknown, paused and maintenance time. A pause requested by an operator is different from a probe failing to submit a result. Scheduled maintenance may be excluded from a contractual calculation, but excluding it should be visible and consistent with the agreed measurement policy.

Do not count a successful request as a minute of availability unless the calculation actually assigns that observation a bounded time interval. Two monitors with different polling intervals produce different numbers of checks; simply pooling their successes can overweight the faster monitor. Time-weighted calculations avoid that particular distortion when the interval classification is explicit.

Calculate the error budget for the same window

A 99.9% availability target permits 0.1% unavailable time. For a 30-day window that is 43.2 minutes. A 31-day calendar month produces a different allowance. Use the uptime calculator to translate a percentage into minutes, and the error budget calculator to explore consumption against a defined target.

An error budget is an operational convention, not proof of a contractual SLA. A contract may use a different window, exclusions or measurement location. Missing observations should not silently replenish the budget. Publish the unknown duration beside the availability figure so a reader can judge the evidence.

Review a real incident timeline

For an outage beginning between two polls, the exact start is uncertain. Preserve the last success, first failure, confirmation results and first recovery. The resulting detection window is more honest than claiming millisecond precision from minute-level checks. Notifications also have their own processing and delivery delay; their timestamps are not the same as the start of the outage.

In UptimeMonitor360, check results, incident evidence and aggregated history serve different purposes. Raw retention can be shorter than aggregate retention. Export evidence while it is still available when you need a detailed postmortem. See the burn-rate guide for deciding when budget consumption needs action.

What to include in a client report

• The exact start and end of the reporting period, with timezone.
• Available and unavailable duration, the calculation method and excluded time.
• Unknown duration and monitoring coverage.
• Incident evidence, detection times and the operator's explanation where known.

This lets a customer distinguish a reliable service from an incomplete measurement.