Dashboard Guide

​

Reliability Summary First

Start at Dashboard -> First-Run Reliability Summary (/dashboard/first-run). This view brings reliability outcomes to the top:

• Deadlock/loop signals
• Context-loss signals
• Tool-call failure signals
• Unnecessary tool-call feedback patterns

​

Tasks

Use task pages to inspect full execution cycles in one place. Key views:

• Timeline: ordered trace, tool_call, and loop events
• Tool Calls: status, errors, latency, arguments/results
• Loops: repeated/circular/stuck patterns with severity
• Detection summary: tool matching, failures, context

​

Conversations

Conversation views aggregate reliability across multi-turn user sessions. Key views:

• /context-failures: typed context regressions
• /context-health: overall score + timeline + severity distribution
• /trace-tree: hierarchical trace/tool graph for root-cause analysis

​

Feedback and Unnecessary Tool Patterns

From task detail pages, submit feedback on tool calls as:

• necessary
• unnecessary
• unclear

Then monitor:

• /api/v1/feedback/patterns
• /api/v1/feedback/stats

​

Alerts

Default alerting is reliability-first and centers on:

• execution_loop_detected
• tool_call_failures
• context_loss_detected
• unnecessary_tool_pattern

Signal-based quality alerts remain available as secondary coverage.