OfficeBench Seed 004

This repository is a minimal benchmark task package for a single Office agent task.

Important scope note:

• this repository publishes the task package, environment note, gold answer, and verifier
• it does not publish an official runner or agent scaffold
• any model result tables kept under docs/ are reference-only archival material and are not part of the benchmark specification

It contains only four release components:

1. input/ Solver-facing task materials:
   • description.md
   • moore_threads_ipo_analyst_model_template.xlsx
   • moore_threads_prospectus_20250926.pdf
2. ENVIRONMENT.md Environment guidance for running the task with any agent/runtime.
3. answer/gold_submission.xlsx The reference workbook.
4. verifier/checker.py A deterministic checker that compares a submission workbook against the gold workbook.

Quick Start

Install verifier dependency:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Run the checker:

python3 verifier/checker.py --submission /path/to/submission.xlsx

Or explicitly provide the gold workbook:

python3 verifier/checker.py \
  --submission /path/to/submission.xlsx \
  --gold answer/gold_submission.xlsx

The checker prints JSON with:

• task_pass
• score
• failures_count
• failures
• check_results

Repository Layout

input/
  description.md
  moore_threads_ipo_analyst_model_template.xlsx
  moore_threads_prospectus_20250926.pdf
answer/
  gold_submission.xlsx
verifier/
  checker.py
ENVIRONMENT.md
requirements.txt

Reference-Only Archive

For internal comparison and historical record only:

• Reference leaderboard (2026-04-23)

This file is intentionally non-normative:

• it is not part of the benchmark definition
• it is not an official evaluation harness
• it mixes multiple execution setups and is kept only as a result archive