Symio-ai/legal-damages-calculator

Model Description

Legal Damages Calculator computes maximum recoverable damages for litigation cases. Given case facts, causes of action, jurisdiction, and evidence of harm, it produces a structured DAMAGES LEDGER per count with: statutory basis, damages type (actual/treble/punitive/fees), calculated amount, justification, and case law averages from similar winning dispositions.

Implements the DAMAGES MAXIMIZATION mandate: never understate, always stack all recoverable categories.

Intended Use

• Primary: Generate damages calculations and ledgers for court filings
• Secondary: Estimate case value for settlement negotiations
• Integration: Called during GLACIER Stage 4 to produce damages sections in filings

Task Type

text-generation -- Structured output generation (damages ledger format)

Base Model

meta-llama/Llama-3.1-8B-Instruct -- Instruction-following capability for structured damages output

Training Data

Source	Records	Description
Jury Verdict Reporters	~500K verdicts	Actual damages awards by category and jurisdiction
Statutory Damages Tables	~5K entries	Statutory damages amounts (treble, penalties, fees)
Settlement Data	~200K	Settlement amounts by case type (anonymized)
Appellate Damages Opinions	~100K	Cases discussing damages calculations
Expert Witness Reports	~50K	Economic damages calculations from expert reports

Damages Categories

• Actual/Compensatory: Lost wages, medical bills, property damage, economic loss
• Consequential: Foreseeable downstream losses
• Treble Damages: Where statute authorizes (e.g., FDUTPA, antitrust)
• Punitive Damages: Where recoverable (with F.S. 768.72 leave in FL)
• Statutory Penalties: Per-violation amounts (e.g., TCPA $500-$1500)
• Attorney Fees: Fee-shifting statutes
• Pre/Post-Judgment Interest: Statutory rates by jurisdiction
• Daily Accrual: Continuing damages (per diem pain and suffering)

Output Format

COUNT 1: [Cause of Action]
  Statutory Basis: [statute]
  Actual Damages: $X (justification)
  Consequential: $X (justification)
  Treble: $X (statute authorizing)
  Punitive: $X (if applicable, with leave requirement noted)
  Fees: $X (fee-shifting statute)
  Interest: $X (rate, period)
  SUBTOTAL: $X
  Case Law Average (5+ similar cases, recency-weighted 2x): $X

Benchmark Criteria (90%+ Target)

Metric	Target	Description
Category Identification	>= 93%	Correctly identify all recoverable damages types
Statutory Basis Accuracy	>= 95%	Correct statute for each damages category
Amount Reasonableness	>= 88%	Within 25% of expert calculation
Stacking Completeness	>= 90%	Must not miss recoverable categories
Case Law Average Accuracy	>= 85%	Averages from real comparable verdicts

GLACIER Pipeline Integration

STAGE 2 (Research) --> damages-calculator identifies all statutory damages provisions
STAGE 4 (First Draft) --> produces DAMAGES LEDGER per count in filing
STAGE 5 (WDC #2) --> WDC validates damages calculations are maximal and accurate

Mandate: Every filing includes a DAMAGES LEDGER. Never aggregate across counts. Always stack: actual + consequential + treble + punitive + fees + interest + daily accrual.

Training Configuration

• LoRA fine-tuning (r=16, alpha=32)
• Epochs: 5
• Learning rate: 1e-4
• Max sequence length: 4096
• Hardware: AWS SageMaker ml.g5.12xlarge

Limitations

• Jury verdict data has survivorship bias (settled cases underrepresented)
• Punitive damages are unpredictable; model provides ranges not point estimates
• Economic expert calculations for complex commercial damages exceed model capability
• Does not account for comparative fault reduction (separate calculation needed)

Version History

Version	Date	Notes
v0.1	2026-04-10	Initial model card, repo created