Skip to content

Create an Entity Resolution Check

Step-by-step tutorial for creating an Entity Resolution check that clusters records describing the same entity and verifies they share one distinction value. For what each property means and how the rule scores and clusters records, see the How It Works page.

Permission Required

You need the Author team permission on the datastore to create an Active check, or Drafter to create it as a Draft. See the Permissions page for the full matrix.

Show me how

The app can walk you through this. Click Show me how in the check creation form's header, or press H while it is open, and the Add a Check walkthrough highlights each step while you fill in the real form.

Field reference

The Authored Check Details form is organized in the sections below. Fill them in as you follow the Steps.

Target

Field Required Type Description
Associate with a Check Template Toggle Links the check to a Check Template. Enabling it replaces the Rule Type dropdown with a Template dropdown where you pick an existing template, which then controls the check's properties. When off, you configure everything on this form.
Rule Type Option The validation logic to apply. Select Entity Resolution for this check. Shown only while Associate with a Check Template is off.
Table / File Option The container the check runs against. Locked after the check is saved. Entity Resolution has no Field picker: the evaluated fields come from the Block and Compare Fields below.
Filter Clause Text A SQL WHERE expression that limits the records entering resolution. Rows outside the filter cannot be clustered or reported.
Custom Anomaly Description Toggle Has no effect on this rule. The option replaces the message on Record Anomalies, and Entity Resolution emits a Shape Anomaly only.

Properties

Field Required Type Description
Distinction Field Option The field that must be consistent inside each cluster. A cluster holding more than one value of this field is reported. Accepts Date, Timestamp, Integral, Fractional, String, and Boolean fields.
Block and Compare Fields List The fields that describe the entity. Click Add target field to add an entry; while the list is empty, clicking the section title does the same. Each entry opens the controls listed under Inside a Block and Compare Fields entry, shows the field type and role as chips in its header, and has a Remove button. At least one entry is required.
Composite Match Threshold Percentage The weighted score two records must reach to land in the same cluster. Defaults to 70%. Raise it to cluster more conservatively, lower it to tolerate more variation.

Inside a Block and Compare Fields entry

Each entry starts with the Field picker. The remaining controls appear once a field is chosen, and which ones appear depends on the field's data type and on the choices above them.

Control Shown for Type Description
Field Every entry Option The source field this entry evaluates. The list offers every field of the selected table or file except the Distinction Field. The field's data type decides how it is compared: Integral and Fractional fields are handled as Numeric, Date and Timestamp as Date / Time, and every other type as String.
Field Role Every entry Option Compare (default) contributes weighted evidence to the match score. Block (advanced) is a hard boundary: records with different values in this field are never paired. Choosing Block sets the comparison to Exact Match and hides the weight.
Comparison Type Every entry Option How two values of the field are compared. String fields offer Fuzzy Match (default) and Exact Match. Numeric fields offer Absolute (delta) (default), Relative (%), and Exact Match. Date / Time fields offer Offset (default), Granularity, and Exact Match. Block fields always use Exact Match.
Weight Compare fields Number How much this field counts in the composite score. Defaults to 1. The same number on every field means equal weighting; raise one to make it count more.
Pair Substrings String, Fuzzy Match Checkbox Treats the values as a match when one is part of the other. On by default.
Pair Homophones String, Fuzzy Match Checkbox Treats the values as a match when they sound alike, even if spelled differently. On by default.
Prioritize rare matches String, Fuzzy Match Checkbox Gives a shared uncommon value more influence on the score than a shared common value. Off by default.
Offset Numeric, Absolute or Relative Number or Percentage The tolerance within which two numbers count as equal: a fixed difference for Absolute (delta), a percentage for Relative (%).
Offset (seconds) Date / Time, Offset Number The number of seconds within which two date or time values count as equal.
Granularity Date / Time, Granularity Option The period two values must share to count as equal: Day, Week, Month, or Year.

Ownership

Field Required Type Description
Owner Option The user responsible for the check. Already filled in with the check creator.
Anomaly Assignee Option The user automatically assigned to anomalies the check produces.

Metadata

Field Required Type Description
Description Text A plain-language description of what the check enforces. Supports Markdown formatting: click the field to open the Markdown editor. Click the Apply suggested description button for a suggestion based on the rule type.
Tags Option Tags applied to the check for filtering and organization.
Additional Metadata Key-value Custom key-value pairs, typically links to catalog entries, tickets, or governance records.

No Pass Criteria section

Entity Resolution does not use coverage, so the form shows no Coverage slider. The composite match threshold, not coverage, controls how tolerant the check is.

Let AgentQ fill in the form

When AgentQ is configured for your deployment, the Check Assistant panel in the check form can author the check for you: describe the entity in plain language (for example, "customers with the same name and city should share one customer ID") and it proposes a configuration you can apply to the form. Review the filled-in fields, then Validate and Save as usual.

Steps

Step 1: Select the source datastore from the left menu, then click the Checks tab.

Step 2: Click Add in the top-right corner and select Check from the dropdown. The Authored Check Details form opens.

Step 3: Select Entity Resolution in the Rule Type dropdown.

Step 4: Fill in the required fields and any optional ones you need. Every field is described in the Field reference above.

Step 5: Click Validate to test the rule against the selected data without saving it. A success message confirms the rule can run on that data. If validation fails, see Troubleshooting.

Step 6: Click Save. A success message confirms that the check was created.

Review the clusters before trusting the threshold

The first run is a calibration run. Open the resulting clusters in the Entity Resolution Recipe and check whether records that landed together really are the same entity, then adjust the threshold or the weights. See Scan and Remediate in the Entity Resolution Recipe.