About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
A synthetic sign-in check is a scripted visitor. On a schedule, from outside your network, it opens your application, signs in with a test account and checks that a page it expects appears. It catches a failure that resource monitoring cannot: the server is running, its processor and disk are fine, and yet nobody can sign in.
What it proves is narrow. One account, one path, from certain places, at certain intervals, worked at that moment. It does not prove the application works for every user, and a simple check that sends one request proves less still. This guide sets out what each kind of probe can show, what the Microsoft and Google documentation says about their limits, and how to choose a first journey worth probing.
Why it matters
Most monitoring watches resources: is the server on, how busy is the processor, how full is the disk. The Google SRE book classes those internal measurements as white-box monitoring and separates them from black-box monitoring, which tests behaviour from outside. It says black-box monitoring has the key benefit of "forcing discipline to only nag a human when a problem is both already ongoing and contributing to real symptoms", and that "for not-yet-occurring but imminent problems, black-box monitoring is fairly useless" (Monitoring Distributed Systems, chapter 6). A synthetic check is a black-box check.
The book also makes the point that matters here: a web service can answer with a success code and still serve the wrong thing, and "only end-to-end system tests can detect that you're serving the wrong content".
Here are some ways a server can be up while sign-in fails (our examples, not sourced):
- The identity service or directory the application relies on is unavailable.
- A database connection pool is exhausted, so the login page loads but the credential check hangs.
- A certificate has expired and browsers refuse the connection.
- A change to the application breaks the sign-in form.
- A third-party service, such as a text-message code provider, is down.
A resource monitor would show green on the web server in all of these (our reading). A probe that tries the journey would not.
This is monitoring, not recovery. A probe tells you quickly that something is wrong. Backup, replication, failover and disaster recovery are the things you use afterwards.
Three levels of probe
The documents we read describe probes at different depths. We group them as three levels (our grouping).
Level 1: reachable. The probe opens a connection or requests a page and checks the status code. Microsoft's standard availability test "checks the availability of a website by sending a single request". It also checks the response and, optionally, the certificate and can check for specified content (Application Insights availability tests, updated 28 September 2026). Google's uptime checks send periodic HTTP, HTTPS or TCP requests and can check the response data.
Level 2: reachable and shows the right content. The probe looks for a string in the response. This catches a server that returns a success code with an error page. Microsoft notes that content match is case-sensitive and "Only English characters are supported with content match", which matters for non-English sites.
Level 3: scripted journey. The probe drives a browser or an API through steps, such as loading the sign-in page, entering credentials and checking for a landing page. Google's synthetic monitors are for this. Its synthetic monitoring overview says a monitor runs a script you write: "you can test the login page of your application, the checkout process of your ecommerce store" and similar. The function "must be written in Node.js", and the documentation says you can configure it to interact with a web page by using Puppeteer, a browser-automation library, or with an API (Create a synthetic monitor, updated 12 November 2026).
The limits of the lower levels are stated plainly by Google in its uptime-check page: "Uptime checks don't load page assets or run JavaScript, and the default configuration of an uptime check doesn't include authentication." Its page also says the queries by individual checkers are stateless, "each query is an independent action". That means a plain uptime check cannot carry a signed-in session from one step to the next.
For Microsoft, the standard test is a single request. The page we read presents custom scripted tests differently: it describes custom availability tests using TrackAvailability() as "archived Classic API guidance", and says to use standard tests for current availability monitoring when possible. We did not find a multi-step sign-in test described on the pages we read; if your provider uses Microsoft tooling, ask what it uses for a sign-in journey.
What a sign-in probe proves
When a scripted sign-in passes, it shows that, at that moment and from those locations (our reading):
- DNS, the network path and the certificate allowed a connection.
- The application served its sign-in page.
- The test account's credentials were accepted by whatever the application checks them against.
- The page after sign-in loaded and contained what the script looked for.
That chain crosses several components that resource monitoring cannot see together, so a pass is meaningful.
What it does not prove
- That other users can sign in. One test account can succeed while an account with different permissions or stored data fails (our reading).
- That the whole application works. The probe checks the path you scripted. A broken report page, upload or payment step is outside it.
- That the test passes for real conditions. The probe usually runs from the vendor's locations. A problem affecting only your office network, or a certain phone network, may not appear.
- That data is correct. A page can load with wrong figures.
- That multi-factor sign-in works. If real users must complete a second factor, a test account that skips it tests a different path. We do not recommend weakening the sign-in rules for the probe; that decision belongs with the security owner (our recommendation).
- That the check itself is healthy. A script can break when the page changes, and then it alarms for the wrong reason.
False alarms, retries and locations
A single failed check is weak evidence. Microsoft's documentation has an option to retry on failure and reports a failure only if three successive attempts fail. It adds: "On average, about 80% of failures disappear on retry." It also recommends several locations: "Our minimum number of recommended test locations is five to ensure that you can distinguish problems in your website from network issues."
Applying that (our reading): a good probe alerts only when several attempts or several locations agree. It trades a little speed for far fewer false alarms. The same page says that, without further steps, you may receive only in-portal notifications and not messages through your configured action groups, which is an example of a check that runs but does not reach anyone. Asking who receives the alert is as important as asking whether the check exists.
The security cost of a probe account
A test account is a real account. Treat it as one (our recommendation):
- Give it the least access that lets the journey run, and no access to real customer records.
- Store its password or token in a secrets store rather than in the script text.
- Keep any saved sign-in state out of shared folders. Playwright's documentation warns: "The browser state file may contain sensitive cookies and headers that could be used to impersonate you or your test account."
- Plan for lockout, and keep the protections on. Microsoft's documentation says that "The Account lockout threshold policy setting determines the number of failed sign-in attempts that will cause a user account to be locked", and that a locked account "can't be used until you reset it or until the number of minutes specified by the Account lockout duration policy setting expires". A probe that runs every few minutes from several locations can add up to many failed attempts, and a locked probe account becomes a monitoring blind spot. Design the probe so that it does not fail sign-in repeatedly, agree with the security owner how its failures are recorded, and do not switch off lockout, multi-factor requirements or security logging to keep the probe working; change the probe instead.
- Agree who owns it, who rotates its credentials, and what happens when someone leaves.
There is a network side too. Microsoft's documentation says its availability test service "uses shared IP addresses, which can expose your firewall-protected endpoints to traffic from other tests", and advises: "Add custom headers to verify the origin of each web request." For endpoints that are not public, it says: "For nonpublic endpoints, monitor an internal health signal from inside the network and create log or metric alerts on that signal." If your application is private, an outside probe may not reach it at all, and you need a different design.
Choosing one journey to probe
This section is our recommendation. Start with one journey, not ten.
- Pick the one the business cannot lose. For many firms that is signing in and opening the main screen, or placing an order.
- Write down what success looks like. A specific heading or value on the landing page, not just a status code.
- Choose an account and an owner. Least privilege, a named owner and a stored secret.
- Set interval and locations. Often enough that the business could act on the alert, from enough places to tell a site fault from a network fault.
- Decide who is told and what they do. A route to a person, and a first action.
- Test the probe safely. Prove the alert route in a way that cannot lock a real account. Use an authorised test account in an isolated test environment, or cause a controlled failure that is not a sign-in attempt, such as pointing the check at a test page that returns an error or looking for text that is not there. Do not use wrong credentials on a production account: repeated failed sign-ins are not harmless, because they can reach the lockout threshold and set off security alerts. Agree the number of attempts and the time with the security owner beforehand, then confirm that the alert arrives and that the probe is restored to its working settings afterwards.
Questions to put to your provider
- Do you run any check that signs in to our application, or only that the server and web page respond?
- Which journey does it test, and which account does it use?
- From how many locations, how often, and how many failures raise an alert?
- Who receives the alert, and through which channel?
- How is the test account secured, and who rotates its credentials?
- How do you detect that the script itself has broken?
- Is multi-factor sign-in part of the tested path?
A worked example (illustrative)
The following is invented. A professional-services firm hosts a client portal. Its server monitoring shows all green on a Monday morning, but clients ring to say they cannot sign in. The cause is an expired certificate on the identity service the portal calls, which the server monitor does not watch. The firm asks the provider for a probe on the main journey: load the portal, sign in with a read-only test account, and check that the document list heading appears. The check runs every five minutes from five locations and alerts after three agreeing failures. The provider documents the account owner and a quarterly credential change. A week later the probe catches a failed deployment within fifteen minutes, before clients notice.
Read next
Related articles you may want next.
- HostingWhat monitoring should a managed hosting customer expect?What a managed hosting customer should ask a provider to define behind 24/7 monitoring: items watched, who is alerted, and what action follows.Read article
- HostingDNS failover: what health-checked DNS can and cannot doWhat health-checked DNS failover can do, how health-check timing and TTL combine for new connections, and why open sessions and data are not moved by DNS.Read article
- HostingReading an availability figure in a service report: what was measured, from where, and what was left outHow to read a service report's availability percentage: what counts as downtime, where it is measured, how intervals work and what is excluded.Read article
Risks and limits
- The documents we read describe Microsoft and Google tools. Other products differ, and the pages change.
- A passing probe does not prove the wider application works.
- A probe account is a security responsibility, and a poorly managed one adds risk.
- Interval, retries and location counts trade cost, speed and false alarms. We give no recommended values.
- We did not test any probe.
Next step
Define one user journey worth probing: the sign-in, the page that should appear and the person who is told if it fails. If you are reviewing a managed service, the Managed Services page is a starting point for that conversation; ask for the scope in writing.
Sources
- Application Insights availability tests, Azure Monitor, Microsoft Learn, last updated 28 September 2026.
- Create public uptime checks, Cloud Monitoring, Google Cloud, last updated 13 November 2026.
- Synthetic monitoring overview, Cloud Monitoring, Google Cloud, last updated 10 November 2026.
- Create a synthetic monitor, Cloud Monitoring, Google Cloud, last updated 12 November 2026.
- Authentication, Playwright documentation, Microsoft, date unavailable.
- Monitoring Distributed Systems (chapter 6), Site Reliability Engineering, Google, O'Reilly Media, copyright 2017.
- Account lockout threshold, Windows security policy settings, Microsoft Learn, last updated 2 November 2018.
All accessed 15 November 2026, except the account lockout page, accessed 16 November 2026.
Learning notes
- A server can be up while users cannot sign in. Resource monitoring does not see that.
- A synthetic check is a scripted visitor that tests from outside on a schedule.
- A pass proves one account on one path worked. It does not prove everyone can.
- A probe account is a real account and needs an owner and a secret store.
- Choose one journey that matters, and say who is told when it fails.
This article was originally published on .