Skip to content
Business BenchA benchmark for AI agents on business work

Tasks·desk·Spreadsheet

What is still coming on our purchase orders

open-purchase-orders·time budget 1200 s·source

The request

The task request supplied to the adapter, alongside the workspace below. Runtime instructions, authorized tools and skills also affect the system. Containerized runs withhold host-side references; local mode is not an isolation boundary.

Leila needs to know what is still coming on our open purchase orders. Check the PO report against the dock log and save open_po_lines.csv; her email has the vendor changes and how she wants it.

Workspace

  • dock_receiving_log.xlsx
  • email_thread_purchasing.txt
  • item_master.csv
  • open_po_report_2026-09-10.csv

Expected deliverables

  • open_po_lines.csv

Recorded acceptance checks

The table uses check names and outcomes from the frozen campaign ledger. Acceptance requires all required predicates to pass; a predicate may itself contain a tolerance or coverage threshold. Its scope is part of the measurement contract.

Recorded checkTypeRequiredFailed / observed
requested columnscsv_columnsyes0 / 6
lines still opencsv_set_equalyes0 / 6
row countcsv_row_countyes0 / 6
open quantity per linecsv_values_matchyes0 / 6
over-receipt flagcsv_values_matchyes0 / 6
Current task declaration (5 checks)

From task.yaml. Original runner grading and frozen retrospective scoring are separate. Read the frozen source for the rules behind the reported results.

  • requested columns — csv_columns, required
  • lines still open — csv_set_equal, required
  • row count — csv_row_count, required
  • open quantity per line — csv_values_match, required
  • over-receipt flag — csv_values_match, required

Recorded results, desk release

pass pass, frozen scorer only fail timed out
SystemRepetitionsFrozen / 3Original / 3Median minutesMean model estimate (USD)
Proto / DeepSeek V4.1 Flash332.10.0265
Codex / GPT-5.6 sol332.00.3237

These are repeated executions on the released fixture, not independent new business cases. Cost is captured model usage priced under the recorded assumptions, excluding infrastructure and human review. Per-attempt check verdicts, raw and frozen verdicts, usage, receipt and artifact hashes for this task are in results/latest/attempts.jsonl under "task": "open-purchase-orders".

Recorded results, desk comparison, 28 September 2026

SystemRepetitionsFrozen / 3Original / 3Median minutes
Proto / GPT-6 sol331.5
Codex / GPT-6 sol330.9
Proto / DeepSeek V4.1 Flash331.0

A separate campaign under the same frozen scorer, not pooled with the release. Conditions and data.