What Is an Evidence Gap in Product Data?

An evidence gap is the difference between what a product record asserts and what its sources can prove. Five states: Known, Missing, Conflicting, Unproven, Stale.

An evidence gap is the difference between what a product record asserts and what its sources can prove. Every field is in one of five states: Known (sourced), Missing, Conflicting, Unproven (asserted without evidence), or Stale (source outdated). A catalog can be 100% filled and still be full of evidence gaps.

Completeness metrics ask whether a field has a value. An evidence gap asks whether that value can be defended. The distinction matters the moment a person other than the data owner, an agent, an auditor or a regulator, reads the record and acts on it. Claro treats every attribute as a claim linked to a source, so the gap is visible per field rather than hidden inside a completeness percentage. See data provenance for how the source link is stored.

The five states, with one industrial example each

Take the certification field on a circuit breaker SKU in a distributor’s ERP:

State	What it means	Example: breaker certification field
Known	Value present and linked to a current source	`UL 489` from the manufacturer's certificate, issued 2025, linked to the PDF
Missing	No value at all	Field empty; the supplier feed never carried it
Conflicting	Two sources disagree	Supplier file says `UL 489`, manufacturer datasheet says `UL 489 and IEC 60947-2`
Unproven	Value present, no evidence behind it	`UL 489` typed in by a buyer years ago; no document linked
Stale	Source exists but is out of date	Certificate expired or superseded by a newer revision of the datasheet

Only the first row can be acted on safely. The second is the obvious problem; the last three look fine in a completeness report, which is why they cause damage.

Filled is not proven

A record with every field populated scores 100% on completeness. If a third of the certification, voltage and safety fields are Unproven or Stale, the record is still not safe to quote, list or let an agent order from. The permission model uses the evidence state as one of its inputs: a Known value may be written automatically, a Conflicting or Unproven one is routed to review or blocked.

Where evidence gaps matter most

Evidence gaps are most expensive on fields with legal, safety or contractual weight: certifications, compliance declarations, hazard data, country of origin. The Battery Evidence Gap Audit applies the five states to battery-containing SKUs, and the confidence thresholds playbook shows how to turn the states into agent permissions.

Glossary

Data provenance

The recorded origin and history behind every attribute.

Glossary

Permission model

What a system may do with each data state.

Playbook

Battery Evidence Gap Audit

Find battery SKUs whose evidence is missing or unproven.

Playbook

Confidence thresholds for AI agent actions

Turn confidence and evidence into permissions.

To find unproven fields in a catalog extract, try the Product Data Completeness Scorer.

FAQ

What is an evidence gap in product data?

An evidence gap is the difference between what a product record asserts and what its sources can prove. A field can be filled and still be an evidence gap if no source backs it, sources disagree, or the source is out of date.

What are the five evidence states?

Known (sourced), Missing (no value), Conflicting (sources disagree), Unproven (a value asserted without evidence) and Stale (the source is outdated). Every field in a product record is in exactly one of these states.

Why is a filled field not the same as a proven field?

A filled field only means someone or something entered a value. A proven field links that value to a document or authoritative source. Agents and regulators act on the record, so an unproven value carries the same risk as a wrong one.

How do you measure an evidence gap?

Count the fields in each of the five states for the SKUs in scope, weighted by action risk. The share of Known fields on certification, compliance and safety attributes is the number to track, not overall completeness.