Skip to main content
For AI agents: a documentation index is available at https://docs.coverbase.com/llms.txt. This page is also available in markdown by appending .md to the URL.
A vendor’s questionnaire answer is a claim. Claim check reads the evidence you already hold on that vendor and says, for each answer, whether the evidence supports it, contradicts it, cannot settle it, or says nothing about it. It runs in the background on a questionnaire submission and appears on the submission’s Claim Check tab. Your Coverbase representative turns on the third-party lifecycle features for your organization.
Claim Check table filtered by verdict, listing five answers marked Contradicted, Needs evidence, Not covered with Answer changed, and Supported, each with a confidence score and the number of passages cited

The Claim Check tab on a submission, with a verdict, confidence and cited evidence for each answer.

​
The four verdicts

Every verdict except Not covered carries the passages it rests on, with the source, the page and the quoted text highlighted.

​
How a verdict is reached

  • The model proposes, code decides. A model reads each answer with the most relevant evidence (between four and ten items) and proposes a verdict with quotes. Code then checks every quote against the source text. A quote that cannot be found in the source is dropped, and a Supported or Contradicted verdict with no quote left falls to Needs evidence, with its confidence capped.
  • Authority comes from the kind of evidence, not the model. Auditor’s reports, accredited certifications, independent assessments, the outside-in scan, licensed ratings and executed contract terms are Authoritative sources. Vendor policies, trust center pages and draft contract terms are Vendor’s own statement. An authoritative contradiction outweighs a vendor’s own statement that agrees, so a trust center page cannot clear a SOC 2 exception.
  • One verdict per answer. Checking again replaces the verdict. When an answer changes after it was checked, the verdict shows Answer changed until it is checked again.
  • Recorded in the Agent Ledger. Every model call is a ledger row, with personal data redaction applied when your organization has it on.

​
What you can do with a verdict

  • Raise finding with this evidence: one finding per question, narrowed to the question and carrying the answer, the verdict and the cited evidence. Answers that already have an open finding are skipped.
  • Ask vendor to clarify: a message on the vendor’s portal request, attached to the question. The vendor sees which question a message is about only when their own form on that request asks it.
  • Check again on selected answers.
Each bulk action reports what it did and what it left alone, for example “3 findings raised, 1 already had a finding”.

​
Where to go next

Claim check guide

Run a check, read the verdicts and act on them.

Due diligence

The rest of the due diligence workbench.

Evidence quality

How source credibility is graded.

Agent Ledger

Where each check is recorded.