Incident Workflow

Reliai is built around a single operational loop: Detect → Understand → Fix → Prove → Share.

Use this guide when an incident is open and you need to move through it systematically.

Detect

Reliai identifies regressions automatically using:

• metric changes (latency, error rate, refusal rate)
• failure pattern clustering across traces
• threshold breaches on tracked signals

When a regression is detected, an incident is opened and signals are grouped into a single investigation unit.

Understand

Use the incident command center to understand what happened.

Start with metrics — identify what changed and when.

Inspect traces — compare failing traces against baseline traces from before the regression.

Review root cause — the root cause panel is computed deterministically from:

• trace comparisons
• prompt or model version changes
• clustering of failure signatures

The root cause is not AI-generated. It is computed from system signals.

You may optionally use the AI explanation to get a plain-language summary of the evidence, but this is grounded in the same deterministic signals.

Fix

Apply a fix based on:

• the recommended action shown in the command center
• your own trace inspection and judgment

Common fixes include prompt changes, model version rollbacks, and guardrail policy updates.

Prove

After applying a fix, Reliai measures resolution impact:

• whether the failure rate declined
• whether metric signals recovered
• how the post-fix traces compare to pre-regression baseline

This is the Fix Verified step. Do not close an incident until you have reviewed the resolution impact.

Share

Export incident context using:

• Ticket draft — an AI-generated draft grounded in the evidence, ready to paste into Jira or GitHub
• Fix summary — a short description of what changed and what was measured

These are AI-assisted drafts. Review before sending.