How Production QA works
Production QA follows your work past the merge. It records every release of a project, collects the errors and alerts your monitoring sends, links each incident to the pull requests most likely behind it, and, where a project has opted in, files a task to fix the problem or revert the change. It is a plan feature: while your plan does not include Production QA, the Incidents page is left out of the side menu, opening it or the page of a release shows Not on your plan (org admins also get See plans), and the project setting reads Off for this org.
Recording and acting
Production QA does two kinds of work, switched on separately.
- Recording runs for every connected project. Releases and incidents are stored and shown on their pages whatever the project's own setting says.
- Acting — filing tasks from incidents, filing reverts and running probes after a release — happens only when the project's Production QA switch is on. That switch is on the project's Production QA tab, and can be turned on only while your plan includes the feature.
Releases
A project connects its releases under Production QA › Connect releases. From then on each release run has its own page: the project and environment, the range of commits it shipped, the pull requests in that range, its status and when it finished. A release moves through Queued, Releasing and then Succeeded, Failed, Cancelled or Skipped. Release workflows (Release pipelines on Azure DevOps) on the project's Production QA tab name the runs that count as releases.
After a production release Jaah watches the project for a short window (the Regression settings: Window (min), Min events, Error rate × and p95 latency ×). If errors or latency climb past those multiples of what they were before, it records a regression. Probes — the health URLs and smoke commands you list — run from a node serving the project after each production release; a failing probe is a health incident.
Incidents
An incident is one problem in production, however many times it happens. It arrives from a monitoring tool connected through an automation source (see Record monitoring alerts as incidents), from a failed release, from a failed probe, from a regression, or from a sustained slowdown of the merge lane. Its Type says which: Monitor, Release failed, Health, Regression or Lane slowdown. Its Severity runs Critical, Error, Warning, Info, Debug.
An incident is Open when first seen, Resolved when it stops, and Reopened if it returns. Each repeat raises its Count and moves Seen forward instead of adding a new row.
Suspects
For each incident Jaah ranks the pull requests most likely to have caused it — usually those in the release that preceded it — and gives each a confidence of High, Medium or Low. The top one shows in the Suspect column. On the incident's Suspects tab a person can Confirm or Dismiss a suspect, Undo that decision, or add one Jaah missed. Every decision shows who made it and when.
From incident to task
With the project's switch on, an incident can become a task. The project's Automations section on the Production QA tab lists the automations that do this, one each for an Incident, a Release failed and a Health breach; Create defaults adds them in one step. At most one open fix task exists per incident, so a noisy alert does not flood the board. The task names the incident and its suspects, and the incident's Tasks tab links back to it.
Revert policy decides what happens when a release causes a regression: Off files nothing, Draft files a draft revert task for someone to start, and Auto-file files one that starts on its own. A revert is an ordinary task: its pull request goes through review and the merge lane like any other, and Jaah never merges it by itself.
Where to see them
Each project's releases are on its pipeline, on the Tasks page: the Release step's Run and each release in the Production section open the page of that release. Incidents sits in the console's side menu for anyone who can view projects, and shows only the projects that person can see. Both pages refresh every 15 seconds.
Open either while your plan leaves Production QA out and it shows only Not on your plan; org admins also get See plans:

How it ties to the rest of Jaah
What happens after a merge feeds each task's Details › Outcome tab and the command center's quality figures, so a change that caused an incident counts against the work that shipped it. See How the command center works, How projects work and How automations work.
To set a project up, see Turn on Production QA for a project.