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

# Inspect source changes

> Investigate source-level changes between a reference run and the selected run.

Investigate source-level changes between a reference run and the selected run.

**MCP tool:** `attribute_drift`. **Operation:** read. Uses the authorized connection; no project or content changes are made by this tool.

## Parameters

| Name | Type | Required | Default | Meaning and accepted values |
| - | - | - | - | - |
| `project_id` | string | Yes | `Required` | Project ID from list\_projects. Select the project in the intended workspace and environment. |
| `run_id` | string | Yes | `Required` | Run ID from list\_runs for this project. Use the exact returned value and check readiness before reading measurements. |
| `reference` | string or object or null | No | `null` | Choose the comparison run: previous\_run, first\_run, or specific\_run. For specific\_run, include run\_id. The request’s run\_id is the checkpoint. |
| `aggregation_level` | string | No | `'domain'` | domain requests source-domain results; url requests individual-page results. Keep the same reporting unit when comparing measurements. |
| `n_boot` | integer | No | `Omit for standard use` | Optional advanced analysis control. Omit for standard reporting; use the advertised runtime schema if an integration requires it. |
| `group_by` | string\[] or null | No | `null` | Split results into rows by the requested dimensions. Check both filters\_applied.group\_by and each row’s group\_by\_values before labeling the groups. |
| `filters` | object\[] or null | No | `null` | Select measured answers with objects containing dimension and values. Use labels from list\_dimension\_options; inspect returned warnings for filters that were not applied. |
| `subject` | string or object or null | No | `null` | Choose sources with kind: leaderboard, all, owned, or specified. For specified, provide domains containing exact source keys for the requested reporting unit. |

## Example call

Replace ID placeholders and example selections with values from your account. This is the `params` object for MCP `tools/call`.

```json theme={null}
{
  "name": "attribute_drift",
  "arguments": {
    "project_id": "PROJECT_ID",
    "run_id": "RUN_ID",
    "reference": {
      "type": "previous_run"
    },
    "aggregation_level": "domain",
    "subject": {
      "kind": "owned"
    }
  }
}
```

## Read the response

Returns source-level contrasts in data.domain\_details. share\_diff compares the checkpoint with the reference; its difference is checkpoint minus reference.

### Fields to use

These selected fields explain the example and the result to preserve. Field presence depends on the response; this is not a complete response schema.

| Field | Meaning and use |
| - | - |
| `data.domain_details[].domain_or_url` | Source whose citation-share change is described. |
| `data.domain_details[].share_diff.value_a` | Reference-run value. |
| `data.domain_details[].share_diff.value_b` | Checkpoint-run value. |
| `data.domain_details[].share_diff.diff` | Checkpoint minus reference. Preserve the scale when displaying the change. |
| `data.domain_details[].share_diff.ci_diff` | Confidence interval on the difference, not on either individual estimate. |
| `data.domain_details[].share_diff.significant` | Returned significance result. It does not establish business impact or causation. |

## Response example

Selected fields from an observed IQRush Dev response. Lists are shortened and project/run IDs are placeholders. Omitted fields remain available in the actual response.

```json theme={null}
{
  "data_ready": true,
  "data": {
    "reference_type": "previous_run",
    "aggregation_level": "domain",
    "domain_details": [
      {
        "domain_or_url": "colorescience.com",
        "share_diff": {
          "value_a": 0.0335164004223577,
          "value_b": 0.026726239624239412,
          "diff": -0.006790160798118287,
          "ci_diff": {
            "lower": -0.019421924882329574,
            "upper": 0.00255965226798446,
            "level": 0.95
          },
          "significant": false
        },
        "is_owned": true
      }
    ]
  }
}
```

Read the actual returned scope, counts, and interval levels before interpreting these fields.

## Important behavior

Use the returned significance and correction metadata. Preserve reference and checkpoint IDs. Do not infer direction from separate confidence bands alone.

Check tool and payload errors before using the result. [Response handling](/developers/guides/09-responses-and-metrics) explains estimates and empty results; [errors and recovery](/developers/guides/13-errors) covers invalid requests and temporary failures.

HTTP equivalent: [Inspect source changes](/developers/http/http-attribute-drift). HTTP parameters and defaults can differ.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.