Skip to main content
A trace span represents an operation in a distributed request. An HTTP call, a database query, a message publish. Tero discovers span patterns from your integrations and builds semantic understanding of what each operation does, how long it takes, and whether it’s worth keeping.

​
Example

Here’s a raw span: 
name: "POST"
span_kind: client
attributes:
  http.method: "POST"
  url.full: "https://api.stripe.com/v1/charges"
  http.status_code: 200
  duration_ms: 234
Tero turns it into a span pattern: 
name: http_call_payment_gateway_charge
service: checkout
span_kind: client
description: Checkout service calling the payment gateway to process a charge.

matchers:
  - field_type: span_attribute
    field_name: http.method
    match_type: exact
    match_value: "POST"
  - field_type: span_attribute
    field_name: url.full
    match_type: starts_with
    match_value: "https://api.stripe.com/v1/charges"
These matchers identify 180,000 spans. Each span has different trace IDs, durations, response codes. But they all match the same pattern: checkout calling Stripe to charge a card. That’s the span pattern.

​
Exploring trace spans

Open any span pattern to see what Tero learned: the description, operation, sample spans, classification reasoning. Browse by service or search across all span patterns. Drill into sample spans to see actual traces. Understand the typical duration, common error codes, which parent spans invoke this operation.

​
Using in chat

Reference a span pattern with @ to focus your questions: 
@postgres_query_select_orders why is this slow?
@http_call_payment_gateway what's the error rate?
@kafka_publish_order_created which services produce this?
Tero pulls in the span’s context: which service owns it, what the operation does, how it connects to other signals. Your question gets answered with that full picture.

​
Improving context

Tero classifies spans automatically, but you know your systems better:
  • Reclassify span patterns that Tero misjudged. That “noise” span might actually matter for debugging.
  • Edit descriptions to add context Tero couldn’t infer.
  • Adjust normalization if an operation is being split into too many patterns or merged incorrectly.
Every correction improves future classifications.