Train verifiers to assess whether NDA documents support legal claims, including cases involving amendments and conflicting versions.

Evaluate whether models can verify agrivoltaic design claims against engineering evidence and provide traceable judgments.

Verify whether microsegmentation policies are correctly reconciled and enforced, even when controller execution encounters complex failure states.

Cross-check endpoint compliance claims against configuration, telemetry, and audit evidence to evaluate evidence reasoning and contradiction detection.

Evaluate how well models identify true food product relationships across multilingual records and highly similar hard negatives.

Evaluate a model's ability to validate specialty intake rules using structured referrals that include both compliant and problematic cases.

Train models to verify affordable housing policy claims against program rules and paired source evidence.

Find workflow dependency errors and analyze their causes with annotated DAG source code.

Test whether models can verify decision-specific capacity assessments against the evidence and identify important gaps and conflicts.

Train models to determine whether evidence truly supports a medical claim using scope-mismatch cases from chronic liver disease literature.

Assess whether psychotherapy model responses faithfully reflect session content with claim-level evidence annotations.

Train and evaluate reliable liability verification by distinguishing allegations in fleet accident filings from findings actually made by a court.

Train precise citation coverage and completeness checks with synthetic social ads and structured evidence annotations.

Train and evaluate ship weight and center-of-gravity calculation checkers with JSON samples containing itemized weights, reference results, and correctness labels.

Verify yield, loss-rate, and mass-balance calculations for ground coffee processing with batch records and reference results.

Evaluate industrial HVAC maintenance answers against procedure references, completeness requirements, and safety restrictions.

Train models to verify buy now, pay later product claims and assess source authority using provider terms and quoted evidence.

The Experiment Design Sketch Image Classification Dataset is used to enhance the automation level of design in industrial manufacturing.

Train models to determine FTTB provisioning eligibility from coverage, resources, port status, and service rules.

Verify restaurant food-handling plans against explicit rules, step by step, across key workflows from receiving to hot holding.