18 record documents, 20,000 rows of raw data
8 pleadings and court documents and 8 CSVs (claims, practitioner rosters, commercial enrollment, receivable ownership and more), all provided as collected with no pre-filtering.
Antitrust Motion to Dismiss: Record Analysis and Memorandum
LAW-ANTITRUST-MTD-001 · Vogler v. Crestline Health Systems · E.D. Tenn.
Prepare the defence's motion to dismiss: define the relevant market, compute shares and HHI, check the statute of limitations from the complaint, three contracts, court orders and 20,000 rows of claims data, and write a memorandum supporting the motion.
AntitrustSherman ActData analysisLegal writingEnglish
1818 files · 20,025 data rows
4.9Mcharacters of material
21scoring criteria
83agent turns in the longest reference rollout
2.29Mcumulative prompt tokens in one rollout
01
Highlights
A built-in privilege trap
Two documents under privileged/ are subject to a Rule 26(b)(5)(B) clawback — the model must recognize this and refuse to use any figure or conclusion from them.
Legal and quantitative reasoning together
Sherman Act §1/§2 elements, relevant markets and submarkets, Rule 12(d) scope, the four-year limitations period and fraudulent-concealment tolling all come down to computable numbers.
Long-horizon tool use
GPT-5.6-sol took 83 turns and 119 tool calls and spawned 3 sub-agents on this task; Qwen3.8-Max hit the 60-minute limit.
02
Instruction excerpt
An excerpt of the instruction the agent actually receives, showing the task setup and constraints.
instruction.md · 28,293 chars
You are an associate at Whitlock, Ainsley & Rourke LLP. The firm represents defendant Crestline Health Systems, Inc. in Vogler et al. v. Crestline Health Systems, Inc., No. 2:24-cv-00419-CTM (E.D. Tenn.). Plaintiffs' First Amended Complaint was filed February 3, 2025. Crestline's responsive motion is due March 17, 2025. Today is March 3, 2025.
The supervising partner needs two things before the motion is drafted: a structured analysis of the record, and a memorandum in support of the motion to dismiss.
How this partner wants a motion worked
Read each document once, straight through, and write your answer down while it is still in front of you. Where the order tells you what rule to apply, apply it — do not go hunting for a second authority that says the same thing. Where two readings are open and the record settles which one is right, settle it and move on. I would rather have your first considered answer on every question in the analysis than your best answer on half of them.
The figures
These are the figures the partner will put in front of the Court, so they must be the figures the record yields, not the figures a spreadsheet yields. The data files are raw. They are published and produced as collected.
Clawed-back documents
The two documents in privileged/ were inadvertently produced and are among the material covered by a pending Rule 26(b)(5)(B) clawback demand. Do not reproduce, quote, or rely on any figure or conclusion from them. The demand is reproduced in the record and identifies what it covers in its own terms; Standing Order ¶ 5 states what follows from it, both for a filing and for a count. Read both.
NOTEExcerpt from the instruction; field definitions and format requirements for both deliverables are omitted. The case is fictional and involves no real parties or medical records.
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 |
|---|---|---|---|
| 01-first-amended-complaint.mdFirst Amended Complaint (numbered paragraphs) | md | 44K chars | 43 KB |
| 02-ppca-blue-ridgeway.mdPPCA · Blue Ridgeway Health Plan | md | 13K chars | 12 KB |
| 03-ppca-volunteer-benefits.mdPPCA · Volunteer Benefits Corp. | md | 15K chars | 14 KB |
| 04-ppca-pinnacle-select.mdPPCA · Pinnacle Select Insurance | md | 13K chars | 12 KB |
| 05-roaa-2018-highland.mdRegional Orthopedic Alliance Agreement (2018) | md | 8,051 chars | 8 KB |
| 06-record-index-and-procedural-history.mdDocket history, tolling agreement, record index | md | 53K chars | 52 KB |
| 07-standing-order-and-local-rules.mdStanding order and local rules | md | 8,507 chars | 8 KB |
| 08-provider-directory-cy2023.mdOrthopedic practitioner directory CY2023 | md | 7,829 chars | 8 KB |
| amp_practice_revenue_cy2018_cy2023.csvPractice revenue by surgeon, year and payer | csv | 672 rows | 48 KB |
| blue_settlement_postings.csvBlue Ridgeway settlement postings | csv | 6,570 rows | 634 KB |
| coe_exception_notices.csvContinuity-of-care notice register | csv | 177 rows | 10 KB |
| commercial_enrollment_cy2023.csvCommercial enrollment by carrier | csv | 10 rows | 497 B |
| orthopedic_claims_adjacent_quarters.csvSupplemental claims for adjacent quarters | csv | 3,316 rows | 1.0 MB |
| orthopedic_claims_cy2023.csvState orthopedic claims CY2023 | csv | 9,225 rows | 2.9 MB |
| privileged/ (2 files)Clawed-back privileged documents (must not be used) | md | 9,245 chars | 9 KB |
| receivable_assignments.csvReceivable ownership and routing | csv | 6 rows | 544 B |
| surgeon_roster_cy2024.csvPractitioner licensure roster | csv | 49 rows | 4 KB |
04
Scoring design
21independent scoring criteria
18 scoring items3 penalty items
Full criteria text, weights and reference answers ship with the complete data package.
21 criteria — 13 binary and 8 five-level Likert — across 8 scoring dimensions.
- 21 criteria: 18 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
Professional standards6
Numerical accuracy4
Instruction following3
Conclusion correctness3
Factual faithfulness2
Structure1
Argument quality1
Compliance1
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 | 31 | 2.29M | 119K | 19m 00s | Completed | |
| GPT-5.6-solClaude Code · Harbor · 3 sub-agents | 119 | 1.72M | 123K | 30m 47s | Completed | |
| Qwen3.8-MaxClaude Code · Harbor | 27 | 1.71M | 114K | 1h 0m | Hit time limit |
06
Specification
| Task ID | LAW-ANTITRUST-MTD-001 |
|---|---|
| Task format | Harbor task.toml · schema 1.4 |
| Package version | 1.4.0 |
| Agent time limit | 1 h |
| Verifier time limit | 5 h |
| Agent container | 2 vCPU · 8 GB memory · 30 GB storage |
| Tool set | filesystem · shell · python |
| Scoring isolation | Separate verifier containerJudge credentials are injected only into the verifier and are invisible to the agent |
| Instruction length | 28,293 characters |
| Case | Fictional; all record documents and data are purpose-built |
| 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