SOURCE-PINNED BENCHMARK RECORDS

Public task library

3,047 public benchmark task definitions and 831 parameterized definitions (templates and level variants) from 11 pinned repositories, with original instructions, verifier references and source snapshots.

Original English task prose is preserved. These are collected definitions, not execution results. Source environments and external assets may need separate setup; source licenses still apply.

WorkArena++ contributes 682 level variants of 341 shared workflows. Parameter configurations and sampled runs do not increase the definition count.

JSONL · Source manifest · Licenses and notices