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

Tasks·desk·Extraction

Key this week's customer POs into order entry

purchase-orders-pdf·time budget 1800 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.

Please get this week's customer purchase orders into the order entry template so we can import them. Carla's note has the details. Save it as po_lines.csv.

Workspace

  • customer_list.csv
  • customer_pos/Northfield_order_NAB-0926.pdf
  • customer_pos/PO PR-24920.pdf
  • customer_pos/Pemberton PO PH-7627 (1).pdf
  • customer_pos/Pemberton PO PH-7627.pdf
  • customer_pos/WPL_PO_4500018241.pdf
  • customer_pos/fax_20260914_0932.pdf
  • note_from_carla.txt
  • order_entry_template.csv

Expected deliverables

  • 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
template columnscsv_columnsyes0 / 6
one row per PO linecsv_set_equalyes0 / 6
row countcsv_row_countyes0 / 6
PO header fieldscsv_values_matchyes0 / 6
part numbers and descriptionscsv_values_matchyes0 / 6
quantities and unit pricescsv_values_matchyes0 / 6
line amountscsv_values_matchyes0 / 6
need-by datescsv_values_matchyes0 / 6
Current task declaration (8 checks)

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

  • template columns — csv_columns, required
  • one row per PO line — csv_set_equal, required
  • row count — csv_row_count, required
  • PO header fields — csv_values_match, required
  • part numbers and descriptions — csv_values_match, required
  • quantities and unit prices — csv_values_match, required
  • line amounts — csv_values_match, required
  • need-by dates — 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 Flash332.00.0403
Codex / GPT-5.6 sol333.20.4344

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": "purchase-orders-pdf".

Recorded results, desk comparison, 28 September 2026

SystemRepetitionsFrozen / 3Original / 3Median minutes
Proto / GPT-6 sol333.5
Codex / GPT-6 sol331.3
Proto / DeepSeek V4.1 Flash331.2

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