Verdicts Explained: UP, DEGRADED, DOWN

The Three Verdicts

UP

Your agent is healthy. Everything is working as expected.

Criteria:

• ≥95% of validations successfully reached your agent
• ≥80% of evaluation prompts passed
• TTFB SLA is met (response within threshold)

What it means: Your users can rely on your agent.

---

DEGRADED

Your agent is working, but there are issues.

Criteria (any of these):

• 20-95% reachability (some validations failing)
• <80% evaluation prompt pass rate (correctness issues)
• TTFB SLA exceeded (slow responses)
• Some regions failing while others succeed

What it means: Your agent mostly works, but some users may experience problems. Investigate soon.

Common causes:

• One region has connectivity issues
• High latency under load
• Intermittent errors
• Partial outage

---

DOWN

Your agent is not working.

Criteria:

• <20% reachability
• Majority of validations failing
• Timeouts or connection errors

What it means: Your users cannot use your agent. Take action immediately.

Common causes:

• Server is down
• Authentication failure (expired token)
• DNS issues
• Firewall blocking requests
• Rate limiting

---

How Verdicts Are Calculated

Agent Status uses threshold-based verdicts, not worst-case. A single failed validation doesn't mean DOWN.

10 validations sent:
- 9 succeed → UP (90% > threshold, still good)
- 7 succeed → DEGRADED (70% < 95%, needs attention)
- 1 succeeds → DOWN (10% < 20%, major problem)

This prevents alert fatigue from transient network blips.

Per-Region Verdicts

Your overall verdict is an aggregate, but Agent Status also calculates per-region verdicts:

Region	Status	Latency
US	UP	245ms
EU	UP	312ms
AP	DEGRADED	4,200ms

This helps identify geographic issues — your agent might be UP overall but DEGRADED in Asia.

Verdict History

Agent Status tracks verdict changes over time:

Feb 25, 08:00 — UP
Feb 25, 09:00 — DEGRADED (AP region slow)
Feb 25, 10:00 — DOWN (API server crashed)
Feb 25, 10:15 — UP (recovered)

This powers:

• Uptime calculations — "99.5% uptime this week"
• Alert triggers — "Status changed from UP to DOWN"
• Incident reports — "Outage lasted 15 minutes"

Special Cases

UNKNOWN

Shown when:

• Agent was just created (no checks yet)
• Agent is paused
• All validations failed due to Agent Status infrastructure issues (not your fault)

REACHABLE (Internal)

Internally, Agent Status distinguishes between:

• REACHABLE — We got a 2xx response, but no semantic verification
• UP — Reachable AND passed evaluation prompts

The dashboard shows UP/DEGRADED/DOWN for simplicity, but API responses include the full verdict.

---

Configuring Thresholds

Currently, verdict thresholds are fixed:

• DOWN: <20% reachability
• DEGRADED: <95% reachability OR <80% eval pass
• UP: ≥95% reachability AND ≥80% eval pass AND SLA pass

Custom thresholds are coming in a future release.