Juricratic
Litigation strategy modeled as a game, not a guess.
Problem
Litigation strategy is mostly built on intuition and war stories. Attorneys make consequential calls about settlement and posture with no model of how the other parties will respond.
Approach
- 01
Model the case as what it structurally is: a multi-party game with payoffs, incomplete information, and sequential moves.
- 02
Compute equilibria over realistic settlement positions, and model co-defendant dynamics where cooperating and defecting have asymmetric costs.
- 03
Ground the payoffs in real court data instead of assumed numbers, so the simulation reflects outcomes that have actually occurred.
What it taught me
The model is only as good as the payoff estimates, and those are where domain expertise is irreplaceable. The valuable output turned out to be sensitivity analysis — which assumptions actually change the recommendation.
Architecture
- Simulation
- Python, game-theoretic solver over party payoffs
- Graph
- Neo4j — parties, claims, precedent relationships
- Orchestration
- LangGraph for multi-step case analysis
- Data
- CourtListener, eyecite for citation extraction