Loading dataset details…
harbor run -d blobfishai/counselbench-100harbor run -d blobfishai/counselbench-100100 high-level legal-agent tasks with 97-asset multi-provider evidence rooms, distinct deep trajectories, and native causal state verification.
harbor run -d blobfishai/counselbench-100100 high-level legal-agent tasks with 97-asset multi-provider evidence rooms, distinct deep trajectories, and native causal state verification.
harbor run -d blobfishai/counselbench-100CounselBench-100 v3.2.5 contains 100 executable legal-agent tasks. Each task has 97 provider-bound assets, twelve evidence-derived portfolio decisions, a closed multi-provider MCP sandbox, and a deterministic causal/state verifier.
harbor download blobfishai/counselbench-100@v3.2.5 --output-dir ./counselbenchThe prompt asks for the employee outcome. A neutral protocol in the evidence room defines the work products without selecting an answer. Gold and verifier state are outside the MCP allowlist. Reference trajectories prove solvability and do not count as leaderboard runs.
CounselBench-100 v3.2.5 contains 100 executable legal-agent tasks. Each task has 97 provider-bound assets, twelve evidence-derived portfolio decisions, a closed multi-provider MCP sandbox, and a deterministic causal/state verifier.
harbor download blobfishai/counselbench-100@v3.2.5 --output-dir ./counselbenchThe prompt asks for the employee outcome. A neutral protocol in the evidence room defines the work products without selecting an answer. Gold and verifier state are outside the MCP allowlist. Reference trajectories prove solvability and do not count as leaderboard runs.