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

# Guardrail Invocations

> Recent guardrail evaluation spans

Recent `tracectrl.guardrail.evaluation` spans for a specific `(agent_id, guardrail_name)` pair, ordered newest first. Reads directly from `otel_traces` and does not depend on the analytics pipeline tick.

### Query Parameters

<ParamField query="agent_id" type="string" required>The agent that owns the guardrail.</ParamField>
<ParamField query="guardrail_name" type="string" required>The guardrail to query.</ParamField>
<ParamField query="limit" type="integer" default="50">Maximum rows to return.</ParamField>

Returns `400` if `agent_id` or `guardrail_name` is missing.

### Response

Array of `GuardrailInvocation` objects:

<ResponseField name="trace_id" type="string">Trace ID of the evaluation span</ResponseField>
<ResponseField name="span_id" type="string">Span ID of the evaluation span</ResponseField>
<ResponseField name="observed_at" type="string">ISO-8601 timestamp</ResponseField>
<ResponseField name="decision" type="string">`pass`, `fail`, or `error`</ResponseField>
<ResponseField name="timing" type="string">`pre_input` or `post_output`</ResponseField>
<ResponseField name="reason" type="string">Judge's reasoning</ResponseField>
<ResponseField name="evidence" type="string">Evidence captured at evaluation time</ResponseField>
<ResponseField name="severity" type="string">Severity level</ResponseField>
<ResponseField name="provider" type="string">`judge_llm` or `protector_plus`</ResponseField>
<ResponseField name="judge_model" type="string">Model used for the decision</ResponseField>
<ResponseField name="response_json" type="string">Structured response payload (Protector Plus only; empty for legacy judge\_llm spans)</ResponseField>