Test whether models can verify experimental facts from evidence in complete virtual laboratory state trajectories.

Connect Arabic expressions to real usage contexts, politeness strategies, and register guidance for more precise retrieval of pragmatic knowledge.

Connect air cargo claims questions to source webpages and specific clauses with traceable evidence for retrieval systems.

Train verifiers to trace cross-document evidence in contract packages and identify whether claims are supported or missing critical grounds.

Train models to interpret varied campus location descriptions and generate accurate delivery tool arguments.

Explore how two-stream workloads and window settings affect join state growth, memory pressure, and output rate.

Retrieve the case file passages behind summary judgment assertions and strengthen evidence-grounded legal search.

Train models to distinguish a clause proven absent from one simply not found in the available material, using contract text and explicit annotations.

Pair commercial construction opportunity notices with structured JSON to train models for accurate field extraction and schema-compliant output.

Transform virtual lab safety notices and procedure requirements into structured rules for accurate safety constraint extraction.

Train models to identify mmWave RAN fault root causes from cross-layer alarms, metrics, configurations, and field observations.

Explore computer and tablet service trajectories covering warranty checks, repair options, service locations, and appointment booking.

Connect source text with supporting passages on metalworking heritage to help models retrieve precise evidence for technique questions.

Train more reliable order-processing agents with tool-use trajectories for computer and tablet order lookup, tracking, and permitted changes.

Connect calibration, readings, units, and significant digits to help models detect inconsistencies in virtual experiment results.

Help models explain traditional theater roles, stage conventions, and performance rules using verifiable source material.

Evaluate how well models interpret delivery terms, acceptance procedures, and party responsibilities in procurement contracts.

Connect technology coverage questions to patent text evidence for grounded patent portfolio question answering.

Connect questions to policy text, supporting quotes, and coverage topics for evidence-grounded collision insurance answers.

Train models to answer French vocabulary questions with clear sense distinctions, natural collocations, and example-based evidence.