HLL 0.1-TX

Humanity's Last Lawsuit

A legal reasoning benchmark built from real Texas appellate opinions decided after every model's training cutoff. We remove the parties, hand the model the facts and procedural history, and ask it to win the case: the dispositive issue, the controlling standard, real authority, application, and the disposition. Graded against what the court actually did. A fabricated citation caps the item at 25%.

See rankings Methodology Submit your model Grades All benchmarks

Live status - real counts, nothing projected

Source opinions

882

full-text SCOTX

Items authored

363

0 invalid schema

Items passing audit

363

verbatim quotes · no leaks

Graded runs

2 model×mode

constitutional · 41civil-litigation · 211other · 5criminal · 25tort · 20real-property · 16family · 33professional-responsibility · 10discovery · 2

Leaderboard see every graded item →

#	Model	Grade	Raw (model alone)	+ DocketRouter model	Lift	Items (raw / model)	Fabricated-cite items (raw / model)
1	anthropic/claude-sonnet-4.5 grades →	F	39%	-	not run yet	34 / 0	17 / 0
2	google/gemini-2.5-flash grades →	F	35%	-	not run yet	1 / 0	0 / 0

How an item is built

1. Real opinion
One published Texas appellate opinion, full text.
2. Blinded
Parties, judges, brands and identifying dates removed. Statutes, rules and other cited cases stay.
3. Gold from the court
Outcome, issue, standard, authorities, application and procedure, each backed by verbatim quotes from the opinion.
4. Audited
Mechanical audit (quotes verbatim, authorities in the text, no name leaks) plus an adversarial reviewer that tries to refute the gold.
5. Graded
Deterministic outcome and authority checks, a three-judge panel for reasoning, and a citation verifier. Judges never see the model name.
6. Raw vs juiced
Every model runs alone and through DocketRouter retrieval. The delta is what our routing is worth.

Coverage plan (targets for the full bank; live counts above are what exists today)

Civil Litigationdifficulty 3-5

Standing, sovereign immunity, TRCP 91a, plea to jurisdiction, removal, summary judgment, appeals

Document-Baseddifficulty 5

Quitclaim deeds, mechanics liens, foreclosure, business acquisition contracts, wills, commercial leases

Court Etiquette / Professional Responsibilitydifficulty 3-5

Ex parte, withdrawal, contempt, sanctions, recusal, ethics, conflict of interest

Discovery Rulesdifficulty 3-5

Spoliation, privilege log, experts, proportionality, RFAs, depositions, financial condition

Personal Injury / Tortdifficulty 3-4

Med mal limitations, premises liability, product liability, GARA, medical expenses, accord & satisfaction

Constitutionaldifficulty 4-5

Ultra vires, waiver of appeal, Miranda, enhancement notice, DWI refusal

Criminaldifficulty 3-5

Traffic stop extension, character evidence, ineffective assistance, Brady, grand jury

Evidencedifficulty 3-4

Hearsay exceptions, expert qualifications, character evidence, business records, impeachment

Family / Divorcedifficulty 3-5

Conservatorship, modification, spousal maintenance, undisclosed property, best interest

Constructiondifficulty 3-5

Mechanics liens, Prompt Pay Act, delay damages, anti-indemnity, unlicensed contractors

Real Propertydifficulty 3-4

Adverse possession, seller disclosure, restrictive covenants, eminent domain, royalties

Business / Contractsdifficulty 3-4

Mandamus, arbitration, parol evidence, temporary injunctions, non-competes

Risk Tolerance / Settlement Strategydifficulty 3-5

TRCP 167, Stowers bad faith, high-low agreements, veil-piercing, preferential setting

Tax / Administrativedifficulty 3-5

Tax deficiency, appraisal protest, unemployment tax, ALJs, agency deference

Edge Casesdifficulty 5

Conflicting authority, first impression, overruled precedent