Service Uptime

Reading this page

The chart can be browsed through earlier history. The summary table stays anchored to the latest published results, showing each rolling period. A tick represents 100% under the calculation below. Select a bar or table result for details.

Connection checks run every 10 seconds; incoming-mail acceptance checks run every 30 seconds from each assigned external path. Results are reported in one-minute intervals. The cached summary is updated every 30 seconds. This page checks it every 10 seconds while visible and updates when the summary changes. Published results typically lag live checks by 2–4 minutes; “Results through” is the finalized result time, not the time your browser refreshed.

Connection checks

Four external testing devices check our public Exchange services every 10 seconds over two independent internet connections. The tests approximate the start of a user connection: they validate TLS and require the expected protocol response or login challenge. Webmail checks follow the OWA sign-in response and verify its expected login form over validated TLS. They do not submit credentials, sign in to an account, access a mailbox, synchronise mail or send a message. Public DNS and authenticated mailbox delivery are not measured by these connection checks.

Incoming mail

MX / Incoming Mail tests the public SMTP service with a small monitoring message and requires acknowledgement after the complete message has been accepted into the mail queue. The test message is then discarded. Each receiving endpoint is tested through two separate external internet paths. One accepting endpoint is sufficient for incoming mail to be available; failures must be confirmed across the receiving endpoints and their independent healthy test paths before a 30-second period is DOWN. Both halves must be DOWN for a DOWN minute.

A degraded incoming-mail result means mail was accepted, but the available onward connection was confirmed unavailable, so delivery may be delayed. These minutes remain available in the uptime percentage and are shown separately in history and incidents. The test proves message acceptance, not delivery to a customer mailbox. Missing test evidence is a monitoring gap, not proof of a service outage.

How a minute counts

Each monitored Exchange server and protocol is assessed separately. One successful connection during a minute makes that minute UP. A minute is DOWN only when both of its 30-second halves contain failed tests from both independent internet connections, with no successful test. A failed test can count only when that tester's separate internet check succeeds.

Short interruptions can therefore occur within a minute counted as UP. This is a measure of availability within each minute, rather than continuous availability every second; sampling and minute boundaries affect which interruptions count. For each displayed protocol, a confirmed DOWN minute on any monitored Exchange server counts once as downtime. Server percentages are not averaged.

Percentages and monitoring gaps

Uptime is 100 × (1 − confirmed outage minutes ÷ total minutes in the selected period). The full period is used: 1,440 minutes for 24 hours, 10,080 for 7 days and 43,200 for 30 days. Recorded minutes with insufficient test evidence remain counted as UP under this policy. More than 60 seconds without a usable test result is identified separately as degraded monitoring; hollow portions of bars mark these monitoring gaps, not confirmed outages. Periods before monitoring began and periods with no minute record are included as assumed UP; this does not mean those minutes were measured. Detailed test logs are retained for investigation.

The graph percentage covers the selected zoom period ending at the right-hand result time; any extra bars to the left do not extend that calculation. The table uses fixed rolling periods ending at the latest result. The monthly view means the preceding 30 days. One DOWN minute in a complete 30-day period gives approximately 99.997685% uptime, displayed as 99.998%. Over 24 hours, one DOWN minute gives 99.931%. Values of at least 99.9995% display as 100%; lower values use three decimal places, omitting an all-zero decimal part.

Your connection may differ

These checks reflect connections from several external test devices. Your ISP, internet routing, firewall or device can experience a problem while the monitors can still connect. A displayed UP result therefore cannot guarantee that every customer can connect at that moment.

For background, see OpenAI's note on aggregate availability and individual experience and Atlassian's minute-based uptime guide. Their calculation rules differ from those described here.