Insights & Actions

The Insights & Actions page is where LakeSentry surfaces optimization opportunities and lets you act on them. Every anomaly, waste finding, and optimization recommendation appears here as an insight with evidence, estimated savings, and — where applicable — an executable action plan.

For the conceptual background on how insights are generated, see Anomaly Detection, Waste Detection, and Action Plans & Automation.

Insights queue

The main view shows all active insights sorted by severity (most critical first), then by most recent. Each row in the table displays:

Column	What it shows
Type	The insight type (e.g., idle_cluster, cost_spike) shown as a badge
Severity	Critical, High, Medium, or Low
Title	Brief description of the finding, with the affected resource name shown below
Savings	Projected monthly savings if addressed (where calculable)
Age	How long ago the insight was created
Action	Action plan status and controls (if an action plan exists)

Filtering

Narrow the queue using the filter controls above the table:

Filter	Options
Search	Free-text search by insight title
Type	Specific insight types (e.g., idle_cluster, cost_spike), with counts shown per type
Category	Anomaly, Waste, Optimization
Severity	Critical, High, Medium, Low (supports multi-select)

Workspace and team filtering is available through the global filter bar in the application header, which applies across all pages including insights.

On the History tab, an additional Status filter is available with options: Auto-Resolved, Action Taken, User Resolved, Dismissed.

Sorting

The table is sorted by severity (most critical first), then by most recent as the default server-side ordering. You can also click column headers to apply client-side sorting within the current page on:

Type — Alphabetical by insight type
Severity — Most critical first
Title — Alphabetical
Savings — Highest dollar impact first
Age — Most recent first

Insight detail

Click any insight to see its full detail page.

Evidence

Every insight includes evidence explaining why it was created:

For anomalies:

Baseline average cost vs. recent cost
Cost multiplier (e.g., 3.2x baseline)
Z-score and confidence level
Number of data points in the baseline

For waste findings:

Type of waste (idle cluster, overprovisioned workers, zombie model, etc.)
Duration of the condition (e.g., “idle for 8 hours”)
Current resource cost and projected waste
Utilization data (CPU, memory, query count)

For optimization suggestions:

Current configuration vs. recommended configuration
Estimated savings from the change
Risk assessment

Timeline

A chronological log of everything that happened with this insight:

When it was detected
Snooze and unsnooze events
Action approval and rejection events
Action execution and failure events
Status change and auto-resolution events
Recurrence detection events

Managing insights

Resolving

Insights can be resolved in several ways:

Resolution	What happens
Auto-resolved	The condition is no longer true (e.g., idle cluster was terminated). LakeSentry resolves it automatically on the next detection cycle.
Resolved by action	An action plan was executed that addressed the issue.
Manually resolved	You mark it as resolved after taking action outside LakeSentry.

Snoozing

Snooze an insight to temporarily hide it from the active queue. Choose a preset snooze duration or enter a custom value:

1 hour
4 hours
24 hours
7 days
30 days
Custom duration (1—720 hours)

Snoozed insights automatically return to the active queue when the snooze period expires. The insight retains its original severity and evidence.

Dismissing

Dismiss an insight if it’s not actionable for your situation. On the insight detail page, this action is labeled Suppress and accepts an optional reason. In bulk operations, the action is labeled Dismiss.

When dismissing or suppressing, you can optionally add a reason explaining why the insight is not actionable.

Dismissed insights are hidden from the active view but accessible through the History tab with the “Dismissed” status filter.

Auto-dismiss rules

Auto-dismiss rules prevent specific insight patterns from cluttering your queue. Each rule can filter on:

Criteria	Description
Insight type	Match a specific insight type (e.g., `idle_cluster`)
Insight category	Match a category (`waste`, `anomaly`, or `optimization`)
Max severity	Only dismiss insights at or below a severity threshold
Max monthly savings	Only dismiss insights with estimated savings below a threshold
Max age	Only dismiss insights older than a specified number of hours
Cluster name pattern	Match clusters whose name matches a pattern (e.g., `dev-%`)
Work unit name pattern	Match work units whose name matches a pattern

Action plans

Insights that have actionable fixes include an action plan section.

Viewing the action plan

The action plan section on the insight detail page shows:

Plan type — What LakeSentry proposes to do (e.g., terminate cluster, resize warehouse)
Safety tier — Tier 0 (autopilot eligible), Tier 1 (approval required), or Tier 2 (manual recommendation)
Estimated savings — Projected cost reduction per month
Status — Current status of the action plan (e.g., pending, approved, executed)

Approving actions

For Tier 1 actions, an admin must explicitly approve:

Review the proposed action and impact assessment.
Check the preconditions (e.g., “cluster is currently idle”).
Click Approve to authorize execution.

Once approved, LakeSentry queues the action for execution by calling the Databricks API. The action plan status on the insight detail page updates to reflect execution progress.

Rejecting actions

Click Reject if you decide not to proceed. Rejected action plans are recorded in the audit log with the rejecting user’s identity.

Autopilot (Tier 0)

Tier 0 actions are classified as safe to auto-execute. These actions have a tier_0 safety tier and may include a cooldown period to prevent repeated execution.

See Action Plans & Automation for details on safety tiers.

Manual recommendations (Tier 2)

Some insights include recommendations that LakeSentry can’t execute automatically. These appear as guidance:

What to do — Step-by-step instructions for the manual change
Where to do it — Link to the relevant Databricks UI page
Expected impact — What should change after you apply the recommendation

Mark the insight as resolved after you’ve taken the recommended action.

Insight history

The History tab shows all resolved and dismissed insights, with:

Column	What it shows
Type	The insight type badge
Severity	Critical, High, Medium, or Low
Title	Brief description, with resolution reason shown below
Resolution	How the insight was resolved: Auto-Resolved, Action Taken, User Resolved, Dismissed, or Superseded
Savings	Realized savings from the resolution
Created	When the insight was first detected
Resolved	When the insight was resolved

The History tab also displays summary cards showing total resolved, total dismissed, realized savings, and average resolution time.

Bulk operations

Select multiple insights using the checkboxes to perform bulk actions:

Dismiss — Dismiss all selected insights (moves them to history)
Resolve — Mark all selected insights as manually resolved
Snooze — Snooze all selected insights for the same duration
Approve — Approve pending action plans for all selected insights (only shown when selected insights have pending actions)
Reject — Reject pending action plans for all selected insights (only shown when selected insights have pending actions)

Next steps

Action Plans & Automation — Safety tiers, guardrails, and automation model
Anomaly Detection — How cost anomalies are detected
Waste Detection & Insights — How waste findings are generated
Audit Log — View the full record of actions taken