Verified Research Reasoning Trajectories: Proof-Process Data for RLVR

Ulam Verified Research Reasoning Trajectories are proof-process data for reinforcement learning with verifiable rewards (RLVR). The dataset contains normalized research problems paired with golden or partial-golden proof graphs. The repository is maintained by ulamai and was last updated on 2026-05-18.

Use Cases

• Training process supervision models based on step-by-step proof trajectories.
• Training proof criticism or judge models based on the golden proof graphs.
• Conducting private evaluations of reasoning agents based on the normalized research problems.
• Reinforcement learning with verifiable rewards (RLVR) based on the proof-process data.

Strengths

• Data is explicitly described as 'verified' and intended for research-level mathematical reasoning.
• Records contain structured proof graphs, which suggests a formal, machine-readable representation.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

ulamai

Freshness

Last updated 2026-05-18 10:46:16; freshness should be verified.