Third-Party Litigation Funding (TPLF) Risk Scoring Engine for Casualty ILS
Third-party litigation funding in jurisdictions like North Carolina is driving legal inflation and impacting casualty insurance-linked securities (ILS). Build an XGBoost risk modeling pipeline combined with natural language court filing parsers to quantify TPLF severity multipliers on casualty claims portfolios.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an ML risk engine combining XGBoost and legal NLP to predict Third-Party Litigation Funding risk in casualty portfolios.
How work is evaluated
Evaluates claim severity multiplier forecasting accuracy and RMSE on test casualty claim filings.
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.
arithmetic_check
Verifies final settlement equals initial reserve times multiplier
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
rmse_multiplier
Root Mean Squared Error of multiplier prediction • target: 0.15 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Extract litigation features (funding state, law firm involvement, venue history) from docket filings
Train an XGBoost regressor model predicting claim tail severity extension
Evaluate casualty ILS portfolio loss reserve impact under varying TPLF penetration rates
Validate model performance against historical loss development triangles
Reference links and supporting material
Casualty claims dataset with jurisdictional court filing metadata and litigation funding tags.
Challenge setup
Evaluates claim severity multiplier forecasting accuracy and RMSE on test casualty claim filings.
Challenge input
JSON containing case venue, injury severity score, and TPLF presence boolean.
XGBoost
Gradient boosted decision trees for tabular risk modeling
Evaluated output
JSON with predicted_severity_multiplier and estimated_final_settlement.
- Verifies final settlement equals initial reserve times multiplier
- Root Mean Squared Error of multiplier prediction • target: 0.15 • range: 0-1
- Benchmark: CasualtyLitigationBench
- Rmse Multiplier target: 0.15
- 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
- Yi-34B · Required
- Google · Optional
- BentoML · Optional
Evaluation contract
- arithmetic_check · Weight 1
- rmse_multiplier · Weight 1
Metrics
- rmse_multiplier · Target 0.15
Benchmarks
- CasualtyLitigationBench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing case venue, injury severity score, and TPLF presence boolean.
Output: JSON with predicted_severity_multiplier and estimated_final_settlement.
Execution harness: Python
Evaluation tools
- XGBoost · Machine Learning Framework
Recipe freeze marker
No recipe freeze time is recorded.