50 real worksheets
Three workbooks — bid cost estimate, two-way cost comparison for projects in progress, and in-process savings — with 50 worksheets in total, not all of them relevant to this task.
Commercial Planning and Margin Analysis for a Bid
ENG-COST-PLAN-001 · Construction cost · before the bid-price review
The bid-price review is tomorrow afternoon. From three real cost workbooks, the model must propose planning options backed by quantities and prices, and deliver a meeting-ready workbook with recomputable formulas plus a briefing note.
ExcelConstruction costCross-sheet joinsRecomputable workpapers
33 files · 50 worksheets
50worksheets
9scoring criteria
64agent turns in the longest reference rollout
4.53Mcumulative prompt tokens in one rollout
01
Highlights
A workpaper that stays useful
The workbook must keep recomputable formulas and include a traceable copy of, or reference to, the base tables so the estimate can be updated after the meeting.
Constraints a program can check
Self-performed margin after planning must equal the margin before planning plus the listed planning amounts, so any pieced-together numbers are exposed.
Very different trajectories
Claude Opus 4.8 finished in 24 turns and 11 minutes; Qwen3.8-Max took 64 turns and 4.53 million prompt tokens; GPT-5.6-sol kept iterating in planning and produced no workbook within the time limit.
02
Instruction excerpt
An excerpt of the instruction the agent actually receives, showing the task setup and constraints.
instruction.md · 790 chars
The bid-price review meeting is tomorrow afternoon. Using the cost estimates in the workspace, prepare a commercial plan and margin analysis that can be presented at the meeting. Please try to finish within 15 minutes.
Propose at least two concrete planning options backed by quantities and prices, and explain their effect on margin and the main implementation risks.
Do not modify the original materials in the workspace. The workpaper should keep recomputable formulas and include a traceable copy of the base tables or external references to them, so the estimate can be updated after the meeting as the bill of quantities and costs change.
Self-performed margin after planning must equal self-performed margin before planning plus the total of the listed planning amounts.
NOTEEnglish translation of the original Chinese instruction; the full instruction is shown except the deliverable list.
03
Attachments
Everything visible inside the agent container, with statistics generated from the task package. Original files ship with the full sample package.
| File | Type | Size | Bytes |
|---|---|---|---|
| 减负信息报表.xlsxIn-process savings and stocktakes: monthly movements, cost analysis summary and stocktake sheets | xlsx | 19 sheets | 627 KB |
| 成本分析报告.xlsxTwo-way cost comparison for projects in progress: labour, materials, plant, subcontracts, measures, fees and taxes | xlsx | 16 sheets | 498 KB |
| 投标成本测算.xlsxBid cost estimate: project overview, quantity analysis, cost summary, labour, subcontracts, materials, overheads, BOQ summary and more | xlsx | 15 sheets | 690 KB |
04
Scoring design
9independent scoring criteria
6 scoring items3 penalty items
Full criteria text, weights and reference answers ship with the complete data package.
9 criteria — 6 five-level Likert positive items plus 3 penalty items, including safe operation and delivery.
- 9 criteria: 6 positive scoring items and 3 penalty items (negate)
- Each criterion is judged separately by an agent judge (individual mode), reproducible and auditable
- Judge credentials are injected only into the verifier; the agent container has no scoring channel
- Full criteria text and reference answers ship with the complete data package
Instruction following2
Numerical accuracy2
Safe operation & delivery2
Professional standards1
Conclusion correctness1
Factual faithfulness1
05
Model results
Trajectory statistics from frontier models in real agent harnesses. One turn is one model call; the context curve shows how state accumulates over a long-horizon task.
| Model | Agent turns | Tool calls | Cumulative prompt tokens | Peak context | Time | Result |
|---|---|---|---|---|---|---|
| Claude Opus 4.8Claude Code · Harbor | 23 | 1.96M | 114K | 10m 38s | Completed | |
| GPT-5.6-solClaude Code · Harbor · 2 sub-agents | 45 | 811K | 117K | 25m 52s | Did not finish | |
| Qwen3.8-MaxClaude Code · Harbor | 63 | 4.53M | 124K | 47m 59s | Completed |
06
Specification
| Task ID | ENG-COST-PLAN-001 |
|---|---|
| Task format | Harbor task.toml · schema 1.4 |
| Package version | 1.0.2 |
| Agent time limit | 20 h |
| Verifier time limit | 5 h |
| Agent container | 2 vCPU · 8 GB memory · 30 GB storage |
| Tool set | filesystem · shell · python · spreadsheet |
| Scoring isolation | Verifier runs as a separate stageJudge credentials are injected only into the verifier and are invisible to the agent |
| Instruction length | 790 characters |
| Data source | Real bid cost workbook (answer-revealing content removed) |
| Reference solution | oracle reward 1.000 |
Contact
Want to run this task on your own platform?
Leave your details and we will send the full sample package for this task (instruction, attachments, verifier and reference trajectories) and walk you through volume and customisation for similar tasks.
- Full task package and verifier
- More tasks in this subdomain
- Integration and re-run support