Casualty Litigation Risk & Nuclear Verdict Forecasting with Google ADK and OpenHands
Nuclear verdicts exceeding $10M increased by 40.7% in recent commercial liability litigations. You will build an automated litigation trajectory forecaster using Google ADK (Agent Development Kit) integrated with Gemini 1.5 Pro and OpenHands code agent execution. The agent must predict nuclear verdict probability with AUC-ROC > 0.85 and forecast damages within 15% error.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an automated legal risk trajectory predictor using Google ADK and OpenHands agent execution to project potential nuclear verdict exposures.
How work is evaluated
Evaluate nuclear verdict probability forecasting accuracy on 200 historical commercial auto liability cases.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
valid_prob_range
Probability must be strictly between 0.0 and 1.0.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
auc_roc
Area under ROC curve for nuclear verdict classification. • target: 0.85 • range: 0.5-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Build a Google ADK agent using Gemini 1.5 Pro to parse legal dockets and claim files.
Integrate OpenHands runtime environment to execute complex statistical regression scripts.
Analyze key nuclear verdict predictors like jurisdiction severity, plaintiff counsel history, and injury type.
Output structured litigation reserve adjustment recommendations for claims officers.
Reference links and supporting material
Dataset of 250 commercial liability court dockets with venue ratings, injury scores, and final settlement/verdict amounts.
Challenge setup
Evaluate nuclear verdict probability forecasting accuracy on 200 historical commercial auto liability cases.
Challenge input
JSON with case_id, venue_county, injury_severity_index, plaintiff_firm_type, initial_demand
Google ADK
Native framework for building structured, multimodal agents powered by Gemini.
OpenHands
Provides dynamic environment execution for code generation and mathematical fitting.
Evaluated output
JSON with nuclear_verdict_probability (0-1), expected_award_usd, recommended_reserve_usd
- Probability must be strictly between 0.0 and 1.0.
- Area under ROC curve for nuclear verdict classification. • target: 0.85 • range: 0.5-1
- Benchmark: NuclearVerdictBench-2026
- Auc Roc target: 0.85
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- OpenHands · Optional
- Google · Required
- OpenHands · Optional
- Google Gemini · Optional
- OpenHands · Optional
Evaluation contract
- valid_prob_range · Weight 1
- auc_roc · Weight 1
Metrics
- auc_roc · Target 0.85
Benchmarks
- NuclearVerdictBench-2026
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with case_id, venue_county, injury_severity_index, plaintiff_firm_type, initial_demand
Output: JSON with nuclear_verdict_probability (0-1), expected_award_usd, recommended_reserve_usd
Execution harness: Python
Evaluation tools
- Google ADK · Agent Framework
- OpenHands · Agent System Tool
Recipe freeze marker
No recipe freeze time is recorded.