russki/agents-last-exam-data

Agents Last Exam — Task Input Data

Input files (the materials each task hands to the agent at run start) for the Agents Last Exam (ALE) benchmark. Browsable per-task directory layout.

The Agents Last Exam dataset family

ALE is published as three companion HuggingFace datasets:

Dataset	Contents	Access
Task Card Metadata	One row per task: titles, prompts, taxonomy, input-file descriptors	Open
Task Input Data	The input/ files each task hands the agent at run start	Open
Reference (Ground-Truth) Data	The reference/ outputs used to score runs	⚠️ Gated (manual approval)

You are viewing: Task Input Data.

• Source repository: https://github.com/rdi-berkeley/agents-last-exam
• Official site: https://agents-last-exam.org

Layout

tasks/
  <task_id>/                # e.g. business_finance/taxform_4_1
    <variant>/              # variant name as defined in the source repo
      input/                # files the agent receives at task start
        ...

What's Included

Only the input/ subdirectory of each task variant. These are the materials the agent is given when a run begins.

What's Excluded

• reference/ — ground-truth outputs. Published separately in the gated reference repo (see the family table above) to protect benchmark integrity.
• software/ — large/licensed software fixtures.
• VM images and any internal artifacts.