How UCF metrics work: real-time coordination monitoring for AI agents

What you see and what it means

The Samsarix dashboard can show six coordination metrics for instrumented runs. But what's actually being measured? And more importantly, what should you do when metrics move?

The calculation pipeline

Data collection

Instrumented actions can emit events such as task started, task completed, handoff, error, retry, and result delivered. Missing instrumentation remains missing evidence; UCF should never imply that an unobserved action was captured.

Metric computation

The UCF engine derives six signals over configured windows. Conceptually, their inputs include:

• Harmony — observed handoff outcomes and coordination consistency
• Friction — retries, errors, timeouts, and other resistance signals
• Throughput — completed work over the selected window
• Focus — evidence that work remained aligned with the selected task
• Resilience — observed recovery behavior after failures
• Velocity — task-completion timing

The deployed endpoint and its versioned implementation are authoritative for exact formulas and windows. These summaries are not a substitute for inspecting the returned fields.

Display

Where live streaming is enabled, the dashboard can receive updates over WebSocket; other surfaces can refresh from the API. Gauges and trends summarize recorded inputs rather than inventing values for unavailable windows.

Reading the signals

Scenario 1: Friction rising, Harmony stable

One possible reading is stable handoffs with rising external resistance. Check API integrations, input format changes, and rate limits before assigning a cause.

Scenario 2: Harmony dropping, Velocity stable

One possible reading is stable execution time with weaker handoff agreement. Check role assignments, overlapping capabilities, and stale shared context.

Scenario 3: Resilience dropping, Throughput stable

Recovery evidence weakened while throughput held. Check error-handling paths, retry policy, and fallback configuration.

Scenario 4: Focus dropping, Harmony stable

Coordination remained stable while task-alignment evidence weakened. Check routing, goal definitions, and agent selection criteria.

Using UCF to debug

When a Spiral fails, UCF can help prioritize where to look:

1. Check Friction first — is the system struggling with inputs?
2. Check Harmony — are agents conflicting?
3. Check Resilience — did the system recover from intermediate failures?
4. Check Focus — were agents working on the right tasks?

Use this as a diagnostic tree alongside logs, traces, provider responses, and the actual task result. Metric movement alone does not prove root cause.

Accessing UCF data via API

UCF metrics are available at:

GET /api/health/detailed
GET /api/coordination/metrics
GET /api/coordination/trends?window=1h

You can feed these into your own dashboards, trigger alerts, or build custom automation rules.

Start monitoring. Try Samsarix free → Inspect UCF signals where the workflow emits sufficient evidence.