Business Evidence Judgment v4
A benchmark that tests whether large language models can correctly determine whether business records support, contradict, or are inconclusive with respect to specific claims, covering six categories of enterprise business scenarios.
Timeline (last 90 days)
The Business Evidence Judgment v4 benchmark tests whether a model can determine whether business records support, contradict, or are inconclusive with respect to a given claim.
The test covers six categories of business scenarios—vendor onboarding, customer delivery, system migration, interface compatibility, configuration release, and data pipelines—comprising 36 claims and producing 108 judgments per model under three source conditions: S0, S1, and S2.
Test condition S0 uses approximately 2,500 words of raw material; S1 adds administrative context to expand to approximately 9,800–9,900 words; S2 adds outdated or similarly scoped distractor records with the same word count as S1.
All Facts (3)
The Business Evidence Judgment v4 benchmark tests whether a model can determine whether business records support, contradict, or are inconclusive with respect to a given claim.
50%UnverifiedThe test covers six categories of business scenarios—vendor onboarding, customer delivery, system migration, interface compatibility, configuration release, and data pipelines—comprising 36 claims and producing 108 judgments per model under three source conditions: S0, S1, and S2.
50%UnverifiedTest condition S0 uses approximately 2,500 words of raw material; S1 adds administrative context to expand to approximately 9,800–9,900 words; S2 adds outdated or similarly scoped distractor records with the same word count as S1.
50%