Evidence & alerts
Read failed steps, download browser artifacts and route monitoring notifications to the right destination.
Start with the saved run for the Check and environment you are investigating. For a PR, select the attempt for the exact head commit. A previous result or a different environment cannot verify the current change.
Read a failure
- Check the target, trigger, version and timestamp. Distinguish a scheduled browser run, a manual run and a PR attempt.
- Read the failed step and assertion or command error. A changed selector, website regression, unreachable preview and runner preparation problem need different fixes.
- Inspect the available screenshots, recording, logs and Playwright trace. Use the evidence to reproduce the issue before changing a test expectation.
- Prepare the fix, validate it locally and verify through the appropriate next run. A manual pass does not resolve a scheduled monitoring incident, and a local pass does not publish a GitHub Check.
For code Checks, dependency installation and command execution have separate output. The report describes those execution steps; it does not reinterpret a test runner's output as a complete list of unit tests.
Download and reuse evidence
The run view lets authorized team members download available screenshots, videos, traces and logs. Artifact links require SiteOS access. Availability depends on the run and capture settings; a missing artifact is not an empty successful recording.
These artifacts come from the check run itself, in Pulse's test browser. The Logs, Traces and Replay pages in Pulse show something else: telemetry and masked recordings that your website and server send from real visits. For an error your visitors hit, start in Errors; it links to a related trace and recording when they exist. See Logs, traces & replay.
Checks that use environment Secrets have stricter diagnostic capture: raw recordings, traces and logs may be withheld to prevent credentials from appearing in saved evidence. PR runs do not receive these Secrets.
For structured data, the CLI provides JSON reads for Check settings, PR attempts and usage:
MCP also provides siteos_pulse_list_runs and siteos_pulse_get_run for saved run history and details. These reads do not start another check. Pulse's downloadable browser artifacts and structured reads are different from a formatted PDF report or a CSV uptime export.
Investigate with your agent
A failed, timed-out or infrastructure-error run offers Fix with AI, which copies a prompt with that run's Check, trigger, target and available evidence. Use AI & MCP for plugin installation and context selection. Give the agent a specific Check or run when you already know which result matters.
Configure notifications
Connect the intended email, Slack or Telegram destination through Integrations, then open Pulse → Notifications. The Monitoring tab routes incidents for the intended Project as an Organization or Personal rule. A provider connection alone does not subscribe every Project or every team member.
Scheduled monitoring confirms a problem after two consecutive eligible failures and resolves it after an eligible scheduled pass. Notification delivery has its own status: an opened incident does not prove a message reached its destination. Recovery notifications follow a successfully delivered failure notification.
PR results are reported through GitHub Checks and the updating PR comment. They do not open or resolve live monitoring incidents. The Pull requests tab can also post to an Organization Slack channel: one thread per pull request with the selected updates, Pulse check results, reviews and comments. Email and Telegram are not available for PR threads. Saving these settings does not send a test or replay earlier PRs. Test both notification delivery and the relevant GitHub round trip when accepting a new setup.
When checks wait
Read the reported reason before retrying. A browser PR can wait for its exact-commit preview; a queued check can wait for runner capacity or the available execution budget. None of those states proves a website failure.
Check Usage & limits, effective schedules and PR eligibility. Follow usage guidance for estimated versus enforced balances. Avoid repeated runs when the saved result already explains the problem.