Governance analytics (dashboard)

The Governance analytics screen helps teams show that controls are working: approval rates, how long human decisions take, and why evaluations were rejected. It complements the row-level Audit log with workspace-level aggregates.

Where to find it

• URL: /dashboard/analytics (authenticated).
• Navigation: sidebar Governance analytics (same session and workspace membership rules as the rest of the dashboard).

Filters

• Workspace — same picker pattern as Audit.
• Period — last 7, 30, 90, or 365 days (based on each evaluation’s created_at).

Data scope

• The page loads up to 2,000 of the most recent evaluations in the selected workspace and time window, then loads related approval stages and audit rows in batches. Very high-volume workspaces may see a capped sample; widen the window or export from the database if you need full history for compliance.

Metrics (definitions)

Metric	Definition
Evaluations	Count of evaluation rows in the workspace and period (subject to the cap above).
Policy pass rate	Among evaluations in a resolved terminal state (auto_approved, approved, rejected, execution_complete), the percentage that ended allowed (auto_approved, approved, or execution_complete).
Human approval rate	Among evaluations that have at least one approval stage and reached a terminal outcome (approved, execution_complete, or rejected), the percentage that ended approved or completed (not human-rejected).
Awaiting human	Count with status pending_human.
Time to first human approval	For human-gated evaluations that ultimately reached approved or execution_complete: elapsed time from evaluation created_at to the earliest stage decided_at where the stage status is approved.
Time to resolution (human-gated)	For human-gated evaluations with a terminal outcome: elapsed time from evaluation created_at to the latest stage decided_at among stages that have a decision timestamp.
Status mix	Counts by evaluation status in the window.
Rejection reasons	For evaluations with status rejected: if an approval_rejected audit event exists → Human rejection; else if approval_timeout_escalation_exhausted → Approval timeout (escalation exhausted); else if approval_timeout with payload.action other than escalate_next → Approval timeout (auto-reject); otherwise the first evaluation_completed audit with engine_outcome no_match or reject → No matching rule or Policy denied (engine); else Other / unknown. See APPROVAL_TIMEOUTS.md.

Median and average times are computed in the app from these deltas (not stored aggregates).

Related

• Audit log — per-evaluation detail and timeline: /dashboard/audit (link from the analytics page).
• Audit data model — evaluations, approval_stages, audit_log (see core migrations and DATA_HANDLING.md).