DATA QUALITY, read 10 Nov 2026
De-duplication is a matching problem on assets and on vulnerabilities, and each product handles one half better than the other
Why the same finding appears twice
A host patched six months late shows up in the authenticated network scan, the agent report and the cloud posture tool, each with its own identifier, severity and first-seen date. Count them as three and the backlog is inflated, the SLA clock starts three times, and three owners may be asked to fix one thing. Count them as one without evidence and a real gap can be hidden.
Correlation works on two keys. The asset key says these records describe the same machine. The vulnerability key says these records describe the same weakness on it. A platform has to resolve both, and failure on either produces duplicates or false merges.
The asset key
Assets are matched on some mix of hostname, fully qualified domain name, IP address, MAC address, cloud instance ID, serial number and agent ID. Each has failure modes: IP addresses are reused by DHCP, hostnames are renamed, cloud instances are replaced under the same name, and containers live for minutes. Ask each candidate which keys it uses, in what order, and what it does on a conflict.
ServiceNow documents the other approach: imported vulnerabilities are reconciled with CMDB assets, and matches create vulnerable items. That makes the CMDB the authority, which is clean if the CMDB is good and a source of unmatched findings if it is not (ServiceNow docs). The guide on what a program needs before a platform covers measuring that overlap first.
The vulnerability key
For infrastructure findings, the natural key is the CVE ID plus the affected software and version on the asset. Scanners mostly agree on CVE IDs but disagree on what they report when a vendor advisory covers many CVEs, or when a detection is a configuration check with no CVE. Application security findings use other keys: a rule ID, a file and line, a package and version. Cloud findings are keyed by resource and control.
A correlation layer needs a mapping between these key types, or it will not merge across source classes. Treat cross-class merging (a dependency scanner and an infrastructure scanner both flagging the same library on a host) as a separate test from same-class merging (two network scanners).
What the vendors state
The statements below are the vendors' own, collected in the facts files on 2026-11-10, and none has been tested by this publication.
- ServiceNow's documentation states that with more than one third-party integration application in use there is no vulnerable item de-duplication across integrations.
- Brinqa describes an AI Deduplication Agent that merges duplicate findings from different scanners into one consolidated, explainable record.
- Hackuity describes automatic de-duplication across all sources and a unified asset inventory with coverage-gap identification.
- Seemplicity states that every finding is de-duplicated, owned and ranked to SLA, and Zscaler's Data Fabric page documents de-duplication across tools and consolidating multiple remediations into one work item.
- Tenable states that ingested third-party data is correlated, normalized and de-duplicated by an ingestion engine, and Nucleus's aggregation page documents asset de-duplication across scanning tools.
A test you can run in a week
Build a small known-overlap set. Pick 50 hosts that two of your scanners both assess, and export the findings from each for the same day. Count the findings in each export and the unique pairs of asset and CVE by hand or with a script. That count is your expected number after de-duplication.
Load both exports into the candidate and compare. Record: the platform's unique count against yours; findings merged that you think are distinct; findings left duplicated; time to first merged view. Then add a deliberate trap. Rename one host, change its IP, and re-import. See whether the platform creates a new asset or follows the old one.
Ask the vendor what happens to the history. When two findings merge, the first-seen date should become the earliest of the two, because SLA math depends on it.
Keep the raw findings
Merged records need to point back to every source record, with each scanner's severity and evidence. Auditors and engineers both ask what the original scanner said, and an engineer will reject a ticket whose evidence cannot be reproduced. Require the platform to show, on a merged finding, the list of contributing sources and the rule that merged them.
Brinqa calls its merged record explainable, which is the right property to ask for from any product. A merged count with no explanation is a number that cannot be defended. Re-run the overlap test after each connector change or platform upgrade, because matching logic changes.
Sources
- ServiceNow documentation, Vulnerability Response import and reconciliation, read 10 Nov 2026
- Brinqa, AI layer (vendor page), read 10 Nov 2026
- Hackuity, risk-based vulnerability management (vendor page), read 10 Nov 2026
- Zscaler, Data Fabric (vendor page), read 10 Nov 2026
- Tenable One (vendor page), read 10 Nov 2026
Last reviewed 10 Nov 2026.