Error Feed Issue Severity and Triage Status

Error Feed severity levels classify how critical each issue is, and status labels track issues through triage from Open to Resolved.

About

Every issue has two independent labels: severity and status. Severity is how bad the problem is. Status is where it sits in your triage workflow. Different purposes, updated independently.

Severity

Severity reflects impact and urgency. It’s assigned automatically based on the error type and quality scores, but you can override it manually on any issue.

SeverityWhat it means
CriticalHigh-confidence, high-impact failure. Likely affecting users now. Examples: safety violations, authentication failures, data exposure, complete task abandonment.
HighSignificant problem with clear user impact. Examples: consistent hallucination, systematic tool misuse, repeated workflow failures.
MediumNotable quality degradation, but not catastrophic. Examples: instruction adherence drift, suboptimal tool choices, inconsistent formatting.
LowMinor issue or edge case. Low frequency or low impact. Examples: slight verbosity, mild instruction drift on an uncommon input.

Severity shows up as a colored badge in the issue list and the issue header. Change it any time from the severity dropdown in the metadata panel.

Tip

Use severity as a prioritization signal. Start every triage session with Critical and High. Low-severity issues are worth tracking but rarely need immediate action.

Status

Status tracks where the issue is in your workflow. Severity describes the problem; status describes what your team has decided to do about it.

StatusMeaning
UnresolvedThe default for any new issue. No one has looked at it yet, or it’s been reviewed but not acted on.
AcknowledgedSomeone has seen this issue and confirmed it’s real. Work may or may not be underway. Use this to signal “we know about it” to the rest of the team.
ResolvedThe underlying problem has been fixed. Error Feed will continue monitoring, and if the same pattern resurfaces it will create a new issue.
EscalatingThe issue is actively getting worse — increasing frequency, expanding impact, or a fix hasn’t held. Use this to flag urgency.

A For Review status is also available from the dropdown for issues that need a closer look before someone decides what to do.

Changing status

Three ways:

  1. Header action buttons: the issue detail header has Resolve, Acknowledge, and Ignore issue buttons for one-click updates.
  2. Status dropdown in the metadata panel: click the current status chip in the Triage section of the right sidebar to switch to any state.
  3. Triage workflow: see Triage Workflow for the full picture, including assigning issues.

Warning

Resolving an issue doesn’t suppress future detection. If the same error pattern shows up again after a fix, it creates a new issue. That’s deliberate: regressions should be visible.

The “first seen” marker

Recently detected issues show a first seen indicator in the feed. This makes it easy to spot new issues without opening every one.

The metadata panel’s Timeline section has the exact first-seen, last-seen, and age (in days) for every issue.

How severity and status interact

They’re independent. An issue can be Critical and Acknowledged (you know it’s bad, you’re working on it), or Low and Escalating (started small but keeps coming up). Set each accurately rather than using one as a proxy for the other.

The feed list is sortable and filterable by both. Typical workflow:

  1. Filter to Unresolved + Critical to find the most urgent uninvestigated issues
  2. Acknowledge issues you’ve reviewed, assign them to the right person
  3. Mark Resolved once the fix is deployed and verified
  4. Watch for the same cluster reappearing — if it does, the fix didn’t hold

See Triage Workflow for step-by-step guidance on working through a batch of issues.

Next Steps

Was this page helpful?

Questions & Discussion