Mitra administrator settings

Settings → Mitra is where an administrator sees and controls Mitra's autonomous operations. It has seven tabs. This page says what each one shows, so you can find the right one and read it correctly. What Mitra is explains the idea; this page is the map of the screens.
Mitra here is a rule engine that watches your records and raises Help Desk cases. It is not a chat assistant and it does not hold conversations.
Mitra Operations: where an administrator sets the rules (a tour)
Before you start
Mitra must be switched on for your subscription. If it is not, opening the page says "Mitra is not enabled for this subscription."
Seeing the tabs needs the permission to view Mitra's autonomous operations. Changing anything needs further permissions, and they are granted to a group by an explicit administrator command, not by an upgrade. In the test workspace used for this page, HR Head could see and configure everything below; HR Ops could see the tabs and change scopes and actions but not the master switch or the policies; and the Manager user group had none of these permissions.
Do not use the buttons described here on a real workspace just to see what happens. Some of them make Mitra act.
The seven tabs
They share one tab row under Settings → MITRA: Autonomous Operations, Policies, Activity, Action Review, Simulation, Engine, Audit.
Autonomous Operations
The control page. It answers "what is Mitra allowed to do here?". The top shows a status line such as "Mitra is watching 9 scopes and may act in 1 of them", then three stages in order: Detection, Recommendation, Execution.
Operational controls: Autonomous execution with Turn execution off, Pause operations (1 hour, 4 hours, until end of day, until manually resumed) and Detection with Stop detecting. Beside them are today's counters: detected today, cases raised, escalations, open situations, minutes since the last run.
Operational scopes: nine cards (Assets, Attendance, Compliance, Employee Records, Help Desk, Leave, Offboarding, Onboarding, Payroll). Each has its own Detection, Recommendation and Execution switches, the number of policies in it, and its open situations.
Autonomous actions: eleven actions grouped as Case management, Communication and follow-up and Resolution. The page states: "An action that is off is never performed, even when an enabled policy asks for it."
Defaults and safety limits: Default Help Desk ticket type, Fallback owner group and Maximum autonomous actions per run, with Save settings.
Explore Mitra: links to Activity, Simulation, Engine and Audit.
See What Mitra is for the checklist to read before turning execution on.
Policies
The list of the situations Mitra watches for. The workspace used for this page had 24, grouped by scope. Each row gives the policy name and version, its Condition in words, its Severity, the Permitted actions, how many Situations it has found, and its State (Enabled, or Not evaluated).
A policy that looks for something being absent will not run while the data it depends on is stale. For example, the policy for a missing attendance record on a working day shows Not evaluated when the data is stale, with the reason: the newest attendance record is older than "the 3 days this policy needs", and missing punches could not be told apart from a feed that had stopped, so nothing was reported. That is the intended behaviour.
Opening a policy shows its condition, severity, whether it raises new cases, its version and who saved it, the actions that would run right now and the ones asked for but blocked (with the reason, for example "This action is switched off"), and then the settings that can be changed: thresholds, follow-up and escalation timing, severity, whether a person must always approve, and who owns the cases it raises. Each change is versioned and asks for a reason. Editing a policy needs the policy-editing permission.
Activity
Every situation Mitra has detected, newest first, with its severity, the policy that found it, who it concerns, how many times it has been seen, its status and the case raised. The filters are Scope, Policy, Status (Detected, Recommended, Acted, Skipped, Resolved, Failed), Action, employee and Seen since. Opening an entry shows the evidence captured at the time and the version of the policy that was in force. It is assembled from stored records and is not recomputed when you open it.
Action Review
Actions Mitra has proposed and is waiting for a person to approve or refuse. The page states the rule: "Mitra proposes; a human decides; the approved action then runs as that human." It has status filters (Open, Proposed, Awaiting approval, Executed, Refused, Failed). When there is nothing to decide it reads "Nothing to review". A person who belongs to a company's Mitra approvers group can see this tab even without the wider permission; it grants sight only.
Simulation
Runs every check and every decision against today's data and changes nothing. You tick the policies to test, or leave all unticked to test every enabled policy, and select Run simulation. The page says: "No ticket is raised, no owner is assigned and no case is closed." The result lists what would be created, escalated and assigned, and to whom. Read it before you allow Mitra to do anything. A simulation was not run while writing this page.
Engine
Whether the scheduled scan is really running. It says it reads the platform's own scheduler, not the app's opinion of it. It shows the state (Running), the schedule (for example every 900 seconds), the last completed run and how long it took, the next run, when the scheduler last started it, the total scheduled runs and the open situations. Below is the Run history, real runs and simulations, each with its trigger, duration, number of policies, situations found, outcome counts (acted, recommended, no change, skipped, failed) and status. Run notes appear under a run, such as a policy that was not evaluated. Run now starts a real run and Simulate opens Simulation.
Audit
Every configuration change and every autonomous decision, with who made it, when, and what changed from and to. The filters are Everything, Configuration changes and What Mitra did. The page says entries are written as they happen and are never edited.
What you should see
Each tab opens on its own address under 
/mitra-operations/, marks itself in the tab row, and states in its subtitle what it is for. Counters and run times change with every scan, so treat any number, including the ones in the captures, as one moment in time.
If it does not work
You see "You do not have permission to manage Mitra's autonomous operations." You lack the view permission. Ask an administrator.
You can read a tab but its buttons do nothing or are missing. Viewing and changing are separate. Changing scopes, actions, policies and the master switch each need their own permission.
A policy says Not evaluated. That is the guard described under Policies, not a fault.
The Engine tab notes that some open detections could not be re-checked because the records they refer to no longer exist. They are left open on purpose.
Not confirmed
Toggling, saving or running anything was not tried live. In particular a simulation and Run now were not run.
The detail pages of an Activity entry and of an Action Review proposal were not tried.
Which other roles can see or change each tab, beyond what the sample user groups showed.
Related
What Mitra is
The Control Center dashboard
Help Desk