Decisions Explorer

The Decisions Explorer is where governance happens at the session level. This guide covers investigation patterns, filter strategies, and workflow integration — how to move efficiently from a compliance alert to the specific sessions that explain it, and how to process large cohorts of sessions for audit evidence.

Reaching the Right Sessions Quickly

Start with the highest-signal filter

When investigating an alert or compliance question, don’t start with a date range — start with the governance signal that matters most:

• Tracking a guardrail regression? Filter by Guardrail Action = blocked + the application, then add a date range.
• Preparing audit evidence for a quarter? Filter by date range first, then optionally narrow by risk level.
• Investigating a specific incident? Search by session ID directly — paste the VeriProof ID or your application’s own trace ID into the search bar for an instant single-session result.

Saved filter combinations

Frequently used filter combinations can be bookmarked using your browser. The Explorer encodes all active filters in the URL, so any filtered view is shareable and bookmarkable.

Investigation Patterns

Pattern 1 — Governance score drop investigation

Scenario: The Analytics Trends view shows a governance score drop starting Tuesday.

1. In the Explorer, set the date range to bracket Tuesday.
2. Sort by Governance score ascending to surface the lowest-scoring sessions first.
3. Scan the sessions — look for a pattern in the Application, Model, or Intent column. If multiple sessions share a common model identifier not seen before Tuesday, a model change is likely the cause.
4. Click the first low-scoring session to open Time Machine. Review the Governance Context tab to see which annotation dimensions contributed to the low score.

Pattern 2 — Pre-audit evidence review

Scenario: An external audit is in two weeks. You need to confirm all HIGH and CRITICAL risk sessions in the last 90 days have been reviewed.

1. Set date range to 90 days.
2. Filter Risk Level = HIGH, CRITICAL.
3. Filter Review status = Pending review.
4. Sort by Timestamp ascending (oldest first — these are most overdue).
5. Work through the queue: click each session, review in Time Machine, submit review.

Alternatively, use Bulk Review if the cohort is large and homogeneous enough to review collectively.

Pattern 3 — New model validation

Scenario: A model was updated in production. You want to confirm governance behavior did not regress.

1. Filter Application = the affected application.
2. Set date range to the 7 days before the model update.
3. Note the governance score distribution.
4. Change the date range to the 7 days after the model update.
5. Compare: did the risk distribution or guardrail hit rate change?

For a rigorous comparison, use the Cohort Comparison view in Analytics instead — it computes statistical difference tests across the two cohorts.

Reading Session Risk Levels

Risk levels are computed by the VeriProof governance pipeline at ingest time. The scale is:

Sessions at HIGH and CRITICAL do not automatically trigger review or alerts — that requires a configured alert rule or review queue rule.

Exporting for External Analysis

For analysis in external tools (BI platforms, audit management software), export the current filtered view:

1. Apply your filters.
2. Click Export → CSV (flat format, all standard fields) or Export → JSON (includes full annotation objects).
3. Large exports (> 10,000 sessions) are queued as background jobs — you receive a notification when the file is ready.

For tamper-evident audit exports, use the Evidence Export workflow in the Application workspace, which produces a signed ZIP archive with a cryptographic manifest.

Related Documentation