| --- |
| license: cc-by-4.0 |
| language: |
| - en |
| task_categories: |
| - question-answering |
| size_categories: |
| - n<1K |
| configs: |
| - config_name: tx |
| data_files: hll-0.1-tx.jsonl |
| tags: |
| - legal |
| - law |
| - benchmark |
| - legal-reasoning |
| - texas |
| - caselaw |
| - evaluation |
| - llm-evaluation |
| pretty_name: Humanity's Last Lawsuit (public split) |
| --- |
| |
| # Humanity's Last Lawsuit — public split |
|
|
| > **Code & tools:** [github.com/docketx](https://github.com/docketx) — [legal-scrambler](https://github.com/docketx/legal-scrambler) pseudonymises a case file on your own hardware before a frontier model sees it; [claude-for-legal](https://github.com/docketx/claude-for-legal) is the Claude Code plugin (`docketx-open-law`) that loads these datasets and checks citations against them. |
|
|
| ## Source & credit — Free Law Project / CourtListener |
|
|
| HLL items are authored against real appellate opinions, and part of that record comes from |
| **CourtListener**. Measured on this public Texas split (115 items): 103 cite an official Texas courts |
| URL as their source of record and **12 cite <https://www.courtlistener.com> directly**; the DocketX |
| opinion corpus the bank is drawn from (`docketx/us-caselaw`, `docketx/texas-caselaw`) is the Free Law |
| Project / CourtListener bulk export of **2026-06-30**. CourtListener is a project of |
| [Free Law Project](https://free.law) — a 501(c)(3) non-profit building free public access to American |
| law since 2010. This benchmark does not exist without that record. |
|
|
| **Free Law Project is a non-profit, and it is worth supporting: <https://free.law/donate/>.** |
| If this data is useful to you, credit Free Law Project and CourtListener too, and consider going |
| straight to the source: CourtListener runs its own |
| [API and MCP server](https://wiki.free.law/c/courtlistener/help/api). |
|
|
| The public split of HLL, a legal-reasoning benchmark built from REAL appellate opinions with the |
| parties redacted. Each item: blinded facts + a question whose gold answer (disposition, controlling |
| standard, authorities) is the court's actual decision — grading is fully mechanical against the |
| source opinion; no model judges another model. The private leaderboard split is withheld. |
|
|
| Per-jurisdiction configs are added as each state's bank matures (Texas first; California in |
| development on the same pipeline). Item provenance: every item carries its source docket, author |
| and critic models, and verbatim evidence quotes from the opinion. |
|
|
| Built by DocketX from the public record — the same team behind the |
| [us-caselaw 51-jurisdiction series](https://huggingface.co/docketx). Site: https://docketrouter.ai/hll |
|
|
