How AI in HR Vault decides what is credible enough to publish.
The vault is not a vendor directory and not an analyst report. Each card is treated as a claim that needs structure, evidence, and human review before it becomes part of the repository.
Editorial rule
Use the submitter's facts. Do not invent outcomes, context, metrics, benchmarks, or business consequences.
Evidence rule
A source must support the specific HR AI implementation, not just the company, vendor, or broad AI topic.
Publishing rule
Cards with major inferred claims, duplicate risk, or unsupported evidence stay in review or are rejected.
Source tiers
S1-S4 evidence labels
The strongest evidence: the named organisation's own website, press release, annual report, regulatory filing, official blog, or named executive interview.
A named third party with institutional credibility, such as a consulting firm, analyst report, industry body, academic source, or named-client vendor case study.
Trade press, business media, or a named individual article that discusses the implementation but is not the originating organisation.
No traceable source, anonymous claim, parked page, unrelated source, or evidence that does not support the specific HR AI use case.
Evidence score
Evidence score is a directional quality signal from 0.0 to 1.0. It combines whether the claim is verifiable, outcome-anchored, and specific enough to falsify.
Review workflow
- 1A practitioner submits a use case with structured taxonomy fields.
- 2Claude rewrites and classifies the submission using facts explicitly provided by the submitter.
- 3Claude validation checks for inferred claims, format issues, taxonomy corrections, and evidence gaps.
- 4Admin review decides whether the card is ready, needs evidence, needs edit, or should be rejected.
- 5Only approved cards appear in the public vault or preview.
Have a real implementation?
Submit one use case to help build the practitioner evidence base.