Guardian Agent Dashboard

Alignment enforced through procedure.

Idle

Current scope

All agents | 2026-02-23 -> 2026-03-08

Overview

Fast answers for owners and stakeholders.

What Stands Out

Current window

Recommended Actions

Most useful next steps

Agent Spotlights

Highest-attention agents in scope

Executive Overview

Agent Leaderboard

Search agents Rows per page

Agent Deep Dive

Use the top selector or click any agent row to inspect trends.

Time Series (chat events)

Date	Chats	Blocked	No Answer	Useful Feedback

Issue Mix

Usage & Access

Admin-only user and question reporting over Q interaction logs.

Admin role required for Usage & Access reporting.

Who Ran Queries / Requests

Amazon Quick query events and Snowflake Agent requests remain explicitly labelled because they are different activity units.

Source	Evidence	Date	User	Agent	Activity Unit	Count	Conversations	Status / Outcomes

Current Q Access

Search access Resource Via

Widest User Access

User	Resources	Agents	Flows	Direct	Group	Owner

Largest Resources

Resource	Type	Users	Direct	Group	Owner

Access Paths

Via	Principal	Users	Resources	Rows

Source	Evidence	User	Resource Type	Resource	Role	Via	Principal	State	Updated

Q Access Changes

Snapshot deltas from Quick metadata syncs. This starts from the first sync after history was enabled.

Source	Evidence	Changed	Type	User	Resource	Via	Role

Guardian API Access

Source	Evidence	Date	Endpoint	User	Role	Status	Requests	Rows

Unusual Behaviour

Investigation leads from transparent rules and evaluated behavioural or semantic vectors. A flag is not a finding of misuse, and blocked or unanswered requests are not proof of an authorization denial.

Source	Inference	Severity	Pattern	User	Score / observed	Comparison	Why flagged	Last Seen

Blocked / Refused / No Answer

Source	Evidence	Time	User	Agent	Status	Prompt	Response

User Activity

Source	Evidence	Period	User	Queries	Agents	Conversations	Agent IDs	Last Seen

Gone Quiet

Compares the selected date range against the immediately preceding range of the same length.

Source	Evidence	User	Baseline Queries	Baseline Agents	Baseline Conversations	Agent IDs	Baseline Last Seen

What People Are Asking

Current Q-log evidence for stakeholder question 8: per-agent outcome volume, common blocked/no-answer patterns, and redacted examples. Semantic top-use-case clustering is not enabled yet.

Source	Evidence	Agent	Outcome	Events	Share	Users	Conversations	Evidence Preview

Snowflake Configured ABAC Scope

Current configured scope only—not complete effective Snowflake access or proof of use. NULL and configured-empty arrays remain distinct.

Source	Evidence	User	Role context	Scope	Masking	Locality	Practice IDs	Data domains	Configuration status	Network aggregate	Source record updated

Snowflake Agent Questions & Outcomes

Curated observability counts with approved redacted samples. Generated SQL remains trace metadata rather than a separate SQL audit source.

Source	Evidence	Date	Agent	Status	Events	Users	Requests	Warnings / Failures	Generated SQL	Evidence Preview

Snowflake Patient / Practice Evidence

Classified observability only. Prompt mentions, trace metadata, and potential access are not proof that exact values were displayed.

Source	Evidence Type	Date	User	Agent	Data Class	Evidence Rows	Patient	NHI	Practice	Redaction

Feedback Explorer

Scope Feedback type Rows per page

			Reason		Conversation	Details

Pipeline Health

Check	Status	Type	Observed and threshold

Ownership Gaps

Scope Rows per page

Ownership & Naming

Admin-only runtime owner mappings and agent display names. Changes apply immediately across the dashboard.

Owner Assignment

No ownership change submitted yet.

Agent Naming

No name change submitted yet.

Agent	Owners	Updated By	Updated At

Guardian Usage Manual

How to configure the dashboard, use each section, and complete common Guardian workflows.

Getting started

Sign in with your Pinnacle account. Guardian starts a new chat by default and either restores your browser snapshot or loads the dashboard from the server.

Sign in. Complete IAM Identity Center sign-in and return to Guardian.
Set the scope. Choose agents and a reporting window with Edit scope.
Refresh. Use the circular-arrow button to load everything available in the selected Normal or Advanced mode for that scope.
Start with Overview. Use Normal mode for the summary or Advanced mode for detailed tables and operational sections.

Header controls

Sign in / Sign out: Starts or ends your Guardian session.
Wing button: Switches between Normal and Advanced presentation without changing your scope or permissions.
Theme button: Cycles through system, light, and dark appearance.
Refresh button: In Normal mode, reloads the stakeholder overview datasets. In Advanced mode, also reloads health, selected-agent detail, and all permitted admin reporting sources.

Status line

The line below the header buttons reports sign-in state, cached data, refresh progress, partial failures, ownership changes, chat activity, and report downloads.

If it says Showing cached dashboard, the displayed snapshot includes its saved time. Press Refresh when you need current server data.

Scope and refresh

The scope bar remains above every section and shows the agents and reporting window currently in use.

Select agents

Choose Edit scope, then search by agent name or runtime ID.
Selecting another result adds it to the current set. Use Clear to return to all agents.
With one agent selected, detail views and reports use that agent. Multiple agents produce an aggregate deep dive.
The selected set is also supplied to Agent Q&A.

Select the reporting window

Use 7D, 14D, 30D, or enter exact From and To dates.
Row limit controls the maximum detail rows returned by applicable requests; it does not cap headline totals.
Date and row-limit edits are saved immediately, but matching mode data loads only after Refresh.
Normal refreshes overview metrics, agents, feedback, and ownership gaps. Advanced also refreshes health and selected-agent detail; admins additionally receive owner mappings and every Amazon Quick, Snowflake Agent, and Guardian API Usage & Access report.
Changing tabs, modes, or Usage & Access source buttons makes no request. These controls only change which already-loaded results are shown.
Normal and Advanced keep separate refresh times. Cached Advanced results survive a Normal refresh only while dates, agents, row limit, and access filters still match; otherwise the affected panels ask for an Advanced refresh.
Refresh never reloads or clears Guardian Chat, and Help is static.
If a selected hosted window has no rows, Guardian can shift to the latest available report date and displays a notice when it does.

Overview

Use Overview to understand volume, reliability signals, feedback, ownership, and which agents need attention.

Normal mode

Stakeholder summary

The opening card summarizes activity for the current scope and window.
Reliability combines blocked and no-answer rates. It is an outcome signal, not a measure that every remaining answer was correct.
User feedback uses explicit feedback events; no feedback is shown as no signal.
What Stands Out, Recommended Actions, and Agent Spotlights highlight the most useful follow-up work.
Select an agent name in a spotlight to move the shared scope to that agent.

Advanced mode

Tables and deep dive

Executive Overview shows active and owned agents, conversations, chat events, blocked and no-answer rates, useful feedback, and owner coverage.
Use Agent Leaderboard search, sortable column headings, row count, and pagination to find an agent.
Select an agent name in the leaderboard to update the shared scope and load Agent Deep Dive.
The deep dive shows chat-event time series and issue mix. Multiple selected agents are aggregated.
Query references are displayed in Advanced mode for traceability.

Usage & Access

This Advanced-mode section is admin-only. It combines several governed reports while preserving the source and evidence type of every result.

All Sources

Use Governance Answers for supported stakeholder questions and Who Ran Queries / Requests for labelled Amazon Quick interactions and Snowflake Agent requests.

Amazon Quick

Review usage, current access, access changes, unusual behaviour, blocked or unanswered interactions, user activity, gone-quiet users, and redacted question examples.

Snowflake Agent

Review request counts, question outcomes, warnings or failures, generated-SQL metadata, and approved patient or practice evidence classifications.

Guardian API

Review endpoint activity and transparent unusual-behaviour flags from Guardian API audit evidence.

Use the reports

Choose a source button to show only the panels supported by that source. This changes presentation; the current scope remains in place.
In Current Q Access, search for a user, resource, group, or principal; optionally choose resource and assignment type; then choose Apply access filter.
Use Widest User Access, Largest Resources, and Access Paths for complete summary slices before inspecting limited detail rows.
Gone Quiet compares the selected window with the immediately preceding window of the same length.
Q Access Changes is derived from metadata snapshots and begins at the first successful baseline after history was enabled.

Read the evidence labels

Observed event / output: The source recorded an interaction, outcome, or classified output.
Observed result access: An approved result or tool output supports the labelled classification.
Prompt mention: The user mentioned something; this does not prove it appeared in returned data.
Potential access: Permissions indicate possible access, not observed use.
Inferred activity: A visible rule derived a flag such as gone quiet or unusual behaviour.
Partial / unsupported: Guardian has incomplete evidence or no approved source for the claim.

Feedback

Use the Advanced-mode Feedback Explorer to inspect feedback users explicitly submitted.

Choose All agents or Selected agents. If no agent is selected, Guardian opens the scope controls.
Choose a Feedback type. The available values come from the feedback loaded for the current dashboard scope.
Choose 10, 25, or 50 rows per page.
Sort by Timestamp, Agent, Type, or Rating, then use Previous and Next to move through results.
Use Reason, Conversation, and Details to understand the submitted event in context.

Health

Health loads with the main Refresh and shows whether the reporting pipeline and supporting metadata are fit for use.

Freshness

Shows whether reports are current and the measured lag in days.

Data quality

Shows the overall result and failed checks. The table lists each check and its type.

Agent mapping

Shows the share of chat events mapped to known agents.

Owner coverage

Shows current coverage against the configured minimum threshold.

Ownership

Use Ownership Gaps to find active agents without an owner. Admins also receive the Ownership & Naming controls.

Review gaps

Choose All gaps or Selected agents.
Sort by Agent, Chats, Blocked, or No Answer.
Choose the page size and use pagination to review the full loaded result.
Selecting an agent elsewhere in Guardian updates the same shared scope used here.

Manage ownership and names — admin

Under Owner Assignment, enter an agent and owner email, choose Add owner or Remove owner, then Apply.
Use selected agent copies the current single-agent scope into the form.
Under Agent Naming, save a friendly display name or clear the existing name.
Changes apply immediately and Guardian refreshes ownership, health, agent detail, and display names.
The mappings table records owners, the updater, and update time.

Agent Q&A and curator reports

Guardian Chat is a read-only analytics assistant. It uses the current agent and date scope and re-checks governed data for every turn.

Ask Guardian

Set the agent and date scope before opening Agent Q&A.
Choose a starter prompt, type a question, or use Ask about this scope to place a scoped prompt in the editor.
Choose Send. Guardian returns the supported answer, citations or query references, and limitations.
Ask follow-up questions in the same conversation. Use New chat to begin a new conversation or Previous chats to continue an older one.

Comparisons are supported; state the periods you want, for example “last 7 days versus the previous 7 days.”

Usage rankings return five results by default. Ask for an exact depth such as “top 7” or “top 12”, then use “show 5 more” to expand the same ranked scope. Chat rankings are capped at 50 identities per source.

Download a curator report

Select exactly one agent in the shared scope.
Set the reporting window and Refresh.
Open Agent Q&A and choose Download a report.
Guardian downloads an HTML report containing summary, what worked, issues, suggestions, and evidence.

The report button is disabled for all-agent or multi-agent scope. Reports are separate artifacts; asking for one in chat does not replace the download action.

Common workflows

These sequences combine the shared scope with the sections described above.

Review an agent

Select one agent and reporting window; Refresh.
Read Overview and its recommended actions.
Switch to Advanced for time series and issue mix.
Review Feedback for submitted reactions.
Ask Guardian a focused follow-up or download the curator report.

Investigate usage and access — admin

Set the relevant agents and dates; Refresh.
Open Usage & Access and choose the source.
Compare observed usage with Current Q Access.
Review access paths, changes, issues, and unusual-behaviour rules.
Record the source, evidence label, and query reference used for the conclusion.

Resolve an ownership gap — admin

Open Ownership and locate the unowned agent.
Select the agent or enter its ID in Owner Assignment.
Add the owner email and apply the change.
Add a friendly name if needed.
Confirm the mappings table and Owner coverage health card update.

Check a suspected data problem

Read the header status and any scope notice.
Confirm agents and dates, then Refresh.
Open Health and check freshness, quality, mapping, and coverage.
Clear agent scope or widen the window to test whether the result is scope-specific.
Use query references when escalating a repeatable problem.

Troubleshooting

Use the status line and current scope first; both are designed to explain the dashboard state.

I cannot see Usage & Access, Feedback, Health, or Ownership: Use the wing-shaped mode button to switch from Normal to Advanced. Usage & Access still requires an admin role.
The dashboard says it is showing cached data: Press Refresh to load current server data. Cached Usage & Access identity and evidence fields may be masked while counts remain available.
I changed dates or the row limit, but the results did not change: Scope controls save the new values immediately. Press Refresh to fetch the complete matching dataset.
Selected-agent Feedback or Ownership is empty: Confirm an agent is selected in Edit scope, Refresh, and then choose Selected agents in the section filter.
A refresh completed with fewer than all panels or health checks: The status line reports partial success. Use the affected section and Health to identify the unavailable source, then retry Refresh.
The report button is disabled: Select exactly one agent. All-agent and multi-agent curator reports are not supported.
Guardian cannot answer a question: Use a specific agent, metric, source, and date range. If Guardian marks the claim partial or unsupported, the approved evidence does not currently prove the full request.
My session expired: Choose Sign in and complete Pinnacle authentication again. Guardian starts a new chat; use Previous chats to load an older conversation.