> ## Documentation Index
> Fetch the complete documentation index at: https://docs.obsy.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Telemetry Review

> AI-powered per-service telemetry quality and cost analysis.

Telemetry Review answers the question: "Is our telemetry actually good?" It analyzes each service's traces, metrics, and logs for quality, completeness, and cost efficiency — and tells you what to fix.

Go to **Telemetry Review** in the sidebar.

***

## What it analyzes

For each service in your catalog, Telemetry Review evaluates:

| Dimension       | What it checks                                                                                   |
| --------------- | ------------------------------------------------------------------------------------------------ |
| **Coverage**    | Are all critical endpoints instrumented? Are there gaps in trace propagation?                    |
| **Quality**     | Are span names meaningful? Are error spans actually marked as errors?                            |
| **Cardinality** | Are there high-cardinality labels (e.g. user IDs in metric labels) that will blow up your costs? |
| **Cost**        | Estimated monthly spend based on current ingestion volume for this service                       |
| **Noise**       | Noisy or useless spans that should be filtered (health checks, heartbeats)                       |

***

## Running a review

1. Go to **Telemetry Review**.
2. Click **Run review** next to a service. You can also run a review from **Service detail → Telemetry tab**.
3. The review runs against the last 24 hours of data from your connected platform (Datadog or Grafana).
4. Results appear within 30–60 seconds.

***

## Reading the results

Each review shows a **score (0–100)** and a set of findings:

### Score bands

| Score  | What it means                                      |
| ------ | -------------------------------------------------- |
| 80–100 | Good telemetry — minor improvements available      |
| 60–79  | Room for improvement — several gaps identified     |
| 40–59  | Significant issues — some areas are blind spots    |
| \< 40  | Poor telemetry — major instrumentation work needed |

### Findings

Each finding has a **severity** (critical / warning / info) and a **recommendation** explaining what to change and why.

Examples:

* `CRITICAL: 40% of spans in checkout-service have no parent span — distributed tracing is broken for this service`
* `WARNING: payment-service.user_id label on metrics has >10,000 unique values — this will cause cost spikes`
* `INFO: /health endpoint is being traced — consider adding it to the drop filter`

***

## Telemetry Standards

Go to **Telemetry Review → Standards** to define what "good" means for your organization:

* Required span attributes (e.g. every span must have `service.version`)
* Required metric labels
* Maximum allowed cardinality per label
* Minimum coverage thresholds per service tier

Standards are evaluated on every review run and violations appear as findings.

***

## Exporting results

Click **Export** on any review result to download a JSON or CSV report. Useful for including telemetry quality in sprint retrospectives or engineering health reviews.