Statistics

Statistics is an operational dashboard for reviewing delivery flow, usage, reliability, and recorded AI cost across workspaces the viewer can access.

Access and filters

Open Management → Statistics; the dashboard requires View Statistics. Non-administrators see only workspaces to which they are currently assigned; administrators can report across the installation. Workspace access is checked on every request.

The workspace and date-range filters apply across all tabs. Date presets are Today, Yesterday, Last 7 Days, Last 30 Days, Last 90 Days, This Month, and Custom (up to 365 days). Results are cached briefly; Refresh bypasses the cache and retrieves current values.

Most metrics are historical for the selected range. Panels explicitly labeled as lifecycle or current-state snapshots show present ticket state and may ignore the date range. Ticket status (All Tickets, Open, Closed) and tag filters apply only to the ticket breakdown charts, not cycle-time or current-state panels. Selecting multiple tags includes tickets carrying any selected tag, and punctuation such as commas remains part of the tag value.

Dashboard tabs

The six tabs separate high-level reporting from detailed operational views.

Tab	Contents
Overview	Summary counts, cost, and prior-period comparisons.
Tickets	Throughput, cycle time, QA, aging, and type/source/priority/assignment breakdowns.
Sessions	Session state, source, model usage, duration, messages, outcomes, skills, subagents, bots, automations, Session Performance, and Guard Events.
User Activity	Active-user counts, messages, ticket activity, review decisions, deployments, and combined activity by user.
More Metrics	Round tables, insights, workflows, and hook-task results.
Cost & Tokens	Recorded spend, tokens, trends, model-level cost, Cache Efficiency, and Budgets.

A tab loads when first opened. Filters invalidate previously loaded tab data so the views remain consistent.

Interpret user activity

User Activity attributes operational actions to the person who performed them within the selected workspace and date range.

• Active Users includes people who create sessions or tickets, contribute round-table turns, make developer or QA decisions, or initiate deployments.
• Developer Actions counts developer approvals and rejections by decision maker.
• QA Actions counts QA approvals and rejections by decision maker.
• Deployments counts deploy-to-slot actions by initiating user.
• Combined Activity totals sessions, tickets created, tickets completed, Developer Actions, QA Actions, and Deployments. Automation authorship is not included.

The summary shows the three action totals above the searchable ranking controls. Selecting a ranking changes the displayed users without changing the workspace or date scope.

Interpret ticket metrics

Ticket metrics characterize delivery flow; use them with the pipeline configuration and date filter visible.

Metric	Interpretation
Completed outcomes	Share of distinct tickets that entered Completed rather than Canceled during the selected period, based on terminal transition time.
Throughput	Distinct tickets with a recorded Completed transition during the selected period, divided by the number of selected days.
Cycle time	Time from ticket creation to its recorded completion in the selected period; the daily trend uses that completion date.
Time to first action	Delay before a ticket first leaves Pending.
Implementation, QA, and PR review time	Time recorded in the corresponding pipeline stage.
QA rejection rate	Frequency with which QA returned work for changes.
Reopened	Tickets that entered Completed more than once.
Open ticket age and blocked age	Current backlog aging, including work that needs intervention.

The Cycle Time Trend uses separate vertical scales: cycle duration in minutes on the left and completed-ticket counts on the right. The legend and tooltip identify each series, and ticket counts are shown as whole numbers.

Stage configuration, manual overrides, and sparse samples affect interpretation. Compare similar workspaces and periods rather than treating one metric as a service-level guarantee.

Interpret session and automation metrics

Session and automation panels identify capacity, reliability, and workload patterns.

• Session status, source, model distribution, daily trend, duration, message count, pull-request rate, outcomes, and high-cost sessions show how work is running.
• Skill usage shows executions, distinct sessions, last-used time, and daily activity using the name recorded when each tool ran. Missing or malformed attribution is omitted.
• Subagent usage shows delegations by subagent name and their daily activity; delegations whose subagent name was not recorded appear as unnamed.
• Bot usage shows conversation volume and frequently used bots.
• Automation metrics show run volume, success rate, duration, cost, and automations with repeated failures or auto-disablement.
• Workflow metrics show run volume, success rate, duration, and frequently used workflows.
• Hook-task metrics highlight failing, retried, or slow lifecycle tasks.
• Session Performance shows duration, response latency, output speed, tool error rate, and interrupt rate, overall and by model.
• Guard Events counts actions the agent's safety guards blocked or warned about, by guard type and model; days without data are shown separately from zero.

Investigate repeated failures in the relevant session, automation, workflow, or hook logs. A dashboard trend identifies where to inspect; it does not replace execution logs.

Cost and token data

Cost panels aggregate usage reported for model interactions and related AI processing. The Overview and Cost & Tokens totals use the same spend scope, including AI merge-conflict resolution; Overview rounds the display to cents.

Translate and Explain usage is included once in these totals and appears as separate workload and module categories. Each action is attributed to the source workspace, so the existing workspace and date filters apply.

Metric	Use
Total cost and daily cost	Track spend for the selected period.
Input, output, cache-read, and cache-write tokens	Understand the recorded usage mix.
Cost per ticket	Compare AI spend associated with completed ticket work.
Cost by model	Identify which selected models account for spend.
Top sessions by cost	Locate sessions for detailed review.
Cache efficiency	Share of prompt tokens served from the provider cache, by workload.
Budgets	Spend against global and workspace limits; select a workspace to see global month-to-date spend.

Values are shown in USD where cost data is available. Missing provider pricing or usage data can produce incomplete totals; reconcile billing with the model provider's invoice. Configure enforcement separately under AI Cost Budgets.

Data scope and retention

Statistics is derived from Polygent records, so report quality follows operational data quality.

• Daily trends follow your browser's time zone, including daylight-saving changes.
• Workspace access is applied before aggregation.
• Deleted or purged records are unavailable for later reporting.
• Current-state panels can change independently of a historical date range.
• Prior-period comparisons use the immediately preceding range of equal duration.
• Recently completed activity may require Refresh before it appears.

Back up the database according to your retention and audit requirements. Export or capture reports externally when long-term evidence must survive application data retention or deletion.

Troubleshooting

Statistics failures are usually caused by access scope, date selection, incomplete source data, or an unhealthy database query.

Symptom	Resolution
Statistics is hidden or access is denied	Grant View Statistics and assign the user to the required workspaces.
A workspace is missing	Confirm workspace membership; administrators see all workspaces.
A panel is empty	Confirm the selected range and workspace contain the relevant records and that the licensed feature is enabled.
Totals appear stale	Select Refresh and verify the source session or ticket has reached the expected state.
Costs do not match an invoice	Check whether all models reported usage and pricing, then reconcile against provider billing.
Custom range is rejected	Select a range no longer than 365 days and ensure the end is not before the start.
One section fails while others load	Retry that section, then inspect system logs and database health if failure persists.
Queries are slow	Narrow the date range or workspace scope, then review database capacity, indexes, and resource pressure.