Symio-ai/legal-element-mapper
0
Symio-ai/legal-element-mapper
Model Description
Legal Element Mapper maps case facts to the required elements of each cause of action. Given a cause of action and the available facts/evidence, it produces a structured element-by-element analysis showing which elements are met, which are weak, and which require additional evidence.
Critical for GLACIER WDC Stage 3 theory validation -- ensuring every cause of action has factual support for every required element before drafting begins.
Intended Use
- Primary: Map facts to legal elements for pre-filing theory validation
- Secondary: Identify evidence gaps and guide discovery strategy
- Integration: Core component of GLACIER Stage 3 WDC "Legal Bench" troop
Task Type
text-generation -- Structured element analysis output
Base Model
meta-llama/Llama-3.1-8B-Instruct -- Strong reasoning for element-by-element analysis
Training Data
Element Mapping Output Format
CAUSE OF ACTION: Negligence (FL)
STATUTE/AUTHORITY: Common law; see Prosser & Keeton on Torts
ELEMENT 1: Duty
Status: MET
Supporting Facts: [fact 1, fact 2]
Supporting Evidence: [exhibit A, testimony B]
Strength: 9/10
ELEMENT 2: Breach
Status: MET
Supporting Facts: [fact 3]
Supporting Evidence: [exhibit C]
Strength: 7/10
ELEMENT 3: Causation (proximate)
Status: WEAK
Supporting Facts: [fact 4 -- circumstantial]
Gap: Need expert testimony on causation chain
Strength: 4/10
Recommendation: Retain expert before filing
ELEMENT 4: Damages
Status: MET
Supporting Facts: [fact 5, fact 6]
Supporting Evidence: [medical records, lost wages docs]
Strength: 8/10
OVERALL: 3/4 elements solidly met. Causation needs strengthening.
RECOMMENDATION: File with current evidence but prioritize causation expert in discovery.Benchmark Criteria (90%+ Target)
GLACIER Pipeline Integration
STAGE 2 (Research) --> statute-parser feeds elements to element-mapper
STAGE 3 (WDC #1) --> element-mapper validates theory: "Are all elements supportable?"
STAGE 4 (Draft) --> element map structures the complaint allegations
STAGE 5 (WDC #2) --> verify each allegation in draft maps to an elementTraining Configuration
- LoRA fine-tuning (r=32, alpha=64)
- Epochs: 5
- Learning rate: 1e-4
- Max sequence length: 8192
- Hardware: AWS SageMaker ml.g5.12xlarge
Limitations
- Complex multi-element causes of action (RICO, antitrust) are more challenging
- Novel causes of action not in training data will produce lower-quality mappings
- Evidence strength scoring is inherently subjective
- Does not replace legal judgment on whether to pursue a particular theory
