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

Tasks·desk·Bookkeeping

West Gallery budget burn for the capital committee

project-budget-burn·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.

Alma needs the West Gallery budget burn for the capital committee - where each cost code stands. Her email explains how she counts it. Save it as burn.xlsx.

Workspace

  • ap_invoices_project_WG-26.csv
  • change_order_log.xlsx
  • email_from_director.txt
  • purchase_orders_west_gallery.csv
  • west_gallery_budget_board_approved.xlsx

Expected deliverables

  • burn.xlsx

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
live formulasxlsx_has_formulasyes0 / 6
no error cellsxlsx_no_errorsyes0 / 6
exhibit fabrication committedxlsx_value_presentyes0 / 6
HVAC remainingxlsx_value_presentyes0 / 6
electrical remainingxlsx_value_presentyes0 / 6
gallery lighting spentxlsx_value_presentyes0 / 6
design fees spentxlsx_value_presentyes0 / 6
total committedxlsx_value_presentyes0 / 6
total remainingxlsx_value_presentyes0 / 6
Current task declaration (9 checks)

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

  • live formulas — xlsx_has_formulas, required
  • no error cells — xlsx_no_errors, required
  • exhibit fabrication committed — xlsx_value_present, required
  • HVAC remaining — xlsx_value_present, required
  • electrical remaining — xlsx_value_present, required
  • gallery lighting spent — xlsx_value_present, required
  • design fees spent — xlsx_value_present, required
  • total committed — xlsx_value_present, required
  • total remaining — xlsx_value_present, required

Recorded results, desk release

pass pass, frozen scorer only fail timed out
SystemRepetitionsFrozen / 3Original / 3Median minutesMean model estimate (USD)
Proto / DeepSeek V4.1 Flash336.30.1054
Codex / GPT-5.6 sol336.10.7655

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": "project-budget-burn".

Recorded results, desk comparison, 28 September 2026

SystemRepetitionsFrozen / 3Original / 3Median minutes
Proto / GPT-6 sol332.9
Codex / GPT-6 sol223.6
Proto / DeepSeek V4.1 Flash331.5

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