The prompt gives no amounts
What to extract, check and compute is left to the model, which must locate the inventory tables, the write-down roll-forward and the by-product and by-ageing detail in 360,000 characters of material.
FINFRAUD-CN-0411 · Quantitative check · STAR Market · FY2023
As a member of an audit engagement team, pull and reconcile figures across the full annual report, the company's written explanation and the auditor's verification notes to re-measure six inventory write-down amounts on the proper basis.
What to extract, check and compute is left to the model, which must locate the inventory tables, the write-down roll-forward and the by-product and by-ageing detail in 360,000 characters of material.
The company's explanation is a management representation; only the procedures and conclusions of the auditor and sponsor count as independent audit evidence.
Each of the six figures needs at least one cross-row, cross-table or cross-period computation, with source and formula shown; a bare conclusion is treated as unverifiable.
Every task in FinFraud-CN-0917 ships with a full Claude Opus 4.8 reference trajectory, and deduplication is zero on all four criteria.
An excerpt of the instruction the agent actually receives, showing the task setup and constraints.
You are a member of an accounting firm's engagement team. XX Semiconductor Technology Co., Ltd. (STAR Market) saw both inventory and its inventory write-down allowance rise sharply in FY2023. The company has explained in writing the composition and ageing of each inventory item, how net realizable value was determined and how the write-down was computed by product model; the annual report notes disclose inventory by category and the movement of the write-down allowance. A special report on the measurement and presentation of the write-down is required, and you are responsible for the quantitative part.
The company's own explanation is a management representation, not independent audit evidence; the procedures and conclusions in the auditor's or sponsor's documents are. All data needed is in these files; this prompt repeats no amounts.
Were the FY2023 write-down charge, its effect on profit and the year-end presentation measured and presented on a proper basis? Re-measured on a proper basis, what are the following amounts?
NOTEEnglish translation of the original Chinese instruction; JSON field names and output structure are omitted.
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 |
|---|---|---|---|
| MATERIALS_INDEX.md.txtMaterials index | md | 379 chars | 829 B |
| annual_report_1.txtFY2023 annual report in full (with financial statements and notes) | txt | 241K chars | 515 KB |
| auditor_reply_1.txtSponsor and auditor verification notes | txt | 57K chars | 126 KB |
| company_reply_1.txtCompany's written explanation (management representation) | txt | 63K chars | 140 KB |
Full criteria text, weights and reference answers ship with the complete data package.
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 · headless, offline | 15 | 692K | 61K | 12m 01s | Completed |
Every task in this dataset ships with a full Claude Opus 4.8 reference trajectory (all tool calls and reasoning); trajectories contain no reference answers or scores.
| Task ID | FINFRAUD-CN-0411 |
|---|---|
| Task format | Harbor task.toml |
| Agent time limit | 30 min |
| Verifier time limit | 5 min |
| Agent container | 1 vCPU · 1 GB memory |
| Network | Fully offline (empty allowlist) |
| Scoring isolation | Separate verifier containerJudge credentials are injected only into the verifier and are invisible to the agent |
| Instruction length | 2,085 characters |
| Dataset | FinFraud-CN-0917 · 1,106 tasks |
| Task type | Quantitative check (553 tasks) |
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.