# Call quality and debugging (/docs/dashboard-guides/call-quality)

> Understand the per-call quality score, replay a call with its event timeline, and read the latency profile.

Every call has a detail view with tools to judge quality and debug what
happened. Open any call from [Call Logs](https://omnidim.io/call-logs),
Omni CRM, a bulk call, or an agent's recent calls, and use the tabs
described below.

## Call Quality Score (CQS) [#call-quality-score-cqs]

CQS is an automatic score from 0 to 5 that rates how well the agent handled
a conversation. It is shown in the call detail's evaluations tab and as a
column in Call Logs and Omni CRM.

The score combines four dimensions:

* **Intent**: did the agent stay on domain, read the caller's intent, and
  handle the input
* **Relevance**: did each reply address the caller, stay factual, and move
  the conversation forward
* **Coherence**: logical flow, clarity, and no needless repetition
* **Latency**: how quickly the agent responded each turn

A short label accompanies the number (Excellent, Good, Average, or Poor).

  CQS runs automatically after each call only when it is enabled for your
  account. You can also recompute it for a single call with the **Regenerate
  CQS** button in the call detail.

## Replay [#replay]

The **Replay** tab plays the recording while the call's events appear in
sync with the audio, turn by turn. It is the fastest way to debug a call:
scrub the audio or click any event to jump to that moment.

* See the conversation as bubbles, including turns that were cut off by the
  caller interrupting (barge-in), and setup events from before the recording
  started.
* Expand any event to view its full details.
* If there is no recording, a silent replay plays the timeline on its own
  clock.

Call traces used for replay are retained for 90 days.

## Latency profile [#latency-profile]

The **Latency Profile** tab breaks down how long each step of the voice
pipeline took, so you can find the slow step before changing anything.

* Summary tiles and statistics for LLM and text-to-speech latency,
  including averages and percentiles
* A per-turn table and charts showing latency for each interaction
* Per-turn speech-to-text (ASR) timing is shown in the interaction detail
  for phone and web calls

Use it to pinpoint whether a slow call is coming from the model, the voice,
or transcription.

## Video walkthrough [#video-walkthrough]