Failure Patterns

The Failure Patterns view (under AI Analysis → Failure Patterns in the sidebar) is a single pane of glass for recurring deployment failures. Rather than re-investigating the same broken step every time it fails on another target, you see one entry per distinct failure signature, ranked by how widely it has recurred.

You need the patterns:read permission to open the view, and patterns:manage to change a pattern’s status.

The dashboard

At the top, four cards summarize your tenant’s recent activity within the detection window:

Active patterns — distinct unresolved failure signatures.
Recurring failures — total failed steps attributed to patterns.
Affected targets — distinct targets that hit a recurring failure.
Alerting — patterns currently above the alert thresholds.

The cluster list

Each row is one pattern, ranked by Deployments (the number of distinct deployments it has appeared in) and then recency — the recurrence-first ordering surfaces the failures that are spreading, not just the noisiest single run. Columns:

Pattern — the AI-assigned title if labeling is enabled, otherwise an Unlabeled (…) fingerprint snippet. An alerting badge marks patterns currently over threshold.
Exit class, Deployments, Occurrences, Targets — the cluster’s blast radius.
Status — active, muted, or resolved.

Use the search box (matches the title and representative error) and the status filter to narrow the list. By default muted patterns are hidden.

Cluster detail

Open a pattern to see:

Aggregate counts and when it was first seen.
AI analysis (if labeling is enabled): the likely cause and step-by-step remediation.
The representative error text for the cluster.
A paginated occurrences table — every individual failure attributed to the pattern, with its deployment, target, solution, and environment.

Muting and resolving

With patterns:manage you can:

Mute — hide the pattern from the default list and stop it alerting. A muted pattern never reopens on its own; use this for known, accepted noise.
Resolve — mark the pattern fixed. Unlike muting, a resolved pattern automatically reopens (“regresses”) to active if the failure happens again — so a fix that didn’t hold resurfaces instead of disappearing.
Reactivate — move a muted or resolved pattern back to active.

Alerts and regressions

When a pattern crosses the configured thresholds (a minimum number of deployments, occurrences, and targets within the alert window), Mantis fires a single emerging-pattern alert through your notification channels — once per crossing, not on every failure. If a resolved pattern recurs, it regresses to active and is eligible to alert again. This is how you learn that a deployment problem is spreading, or that a past fix has broken.

Refreshing

The view live-updates as the background detector clusters new failures. You can also click Recompute now to trigger an immediate pass scoped to your tenant.