Symio-ai/legal-timeline-builder

Model Description

Legal Timeline Builder extracts chronological events from legal documents (complaints, depositions, discovery, correspondence) and constructs a unified case timeline. It identifies dates, events, actors, documents, and causal relationships, producing a structured chronology that maps to causes of action and statute of limitations calculations.

Essential for case organization and statute of limitations analysis in the GLACIER pipeline.

Intended Use

• Primary: Build case timelines from multiple document sources
• Secondary: Statute of limitations analysis (identify triggering events and deadlines)
• Integration: Powers timeline construction throughout the GLACIER pipeline

Task Type

token-classification -- Temporal event extraction and relationship labeling

Base Model

nlpaueb/legal-bert-base-uncased -- Legal domain understanding for date/event extraction

Training Data

Source	Records	Description
Case Chronologies	~100K	Attorney-prepared case timelines
Complaint Timelines	~200K	Factual allegations with dates extracted
Discovery Timelines	~150K	Document-based chronologies
SOL Analysis Documents	~50K	Statute of limitations calculations
Expert-Annotated Documents	~30K	Documents with event/date annotations

Extraction Entities

• DATE -- Explicit date (April 5, 2024) or relative (three weeks later)
• EVENT -- What happened (meeting, transaction, injury, filing)
• ACTOR -- Who was involved (person, entity)
• DOCUMENT -- Referenced document (contract, email, letter)
• CAUSAL_LINK -- "because of", "resulting in", "which caused"
• SOL_TRIGGER -- Event triggering statute of limitations
• SOL_DEADLINE -- Computed filing deadline

Output Format

CASE TIMELINE: [Case Name/Number]

CRITICAL DATES:
  [date]: SOL TRIGGER - [event] (source: [document])
  [date]: SOL DEADLINE - [X years from trigger per statute Y]

CHRONOLOGICAL EVENTS:
  1. [date] | [event] | [actors] | [source document] | [relevant to: Count X]
  2. [date] | [event] | [actors] | [source document] | [relevant to: Count Y]
  ...

SOL ANALYSIS:
  Count 1 ([COA]): Triggered [date], deadline [date], status: TIMELY/EXPIRED
  Count 2 ([COA]): Triggered [date], deadline [date], status: TIMELY/EXPIRED

GAPS:
  [Periods with no documented events that may need investigation]

Benchmark Criteria (90%+ Target)

Metric	Target	Description
Date Extraction F1	>= 95%	Correctly extract all dates
Event Extraction F1	>= 88%	Correctly identify events
Temporal Ordering	>= 93%	Events in correct chronological order
SOL Trigger Identification	>= 92%	Correctly identify SOL-triggering events
Cross-Document Alignment	>= 85%	Same events from different sources correctly merged

GLACIER Pipeline Integration

STAGE 2 (Research) --> timeline-builder constructs initial timeline from complaint
STAGE 3 (WDC #1) --> timeline validates SOL compliance for all causes of action
STAGE 4 (Draft) --> timeline structures factual allegations chronologically
Post-discovery --> timeline updated with deposition and document evidence

Training Configuration

• Epochs: 10
• Learning rate: 3e-5
• Batch size: 16
• Max sequence length: 2048
• Hardware: AWS SageMaker ml.g5.2xlarge

Limitations

• Ambiguous dates ("sometime in early 2023") produce approximate placements
• Events spanning long periods may be misplaced
• Relative dates ("three months after the incident") require correct anchor resolution
• Cross-document event deduplication is challenging when descriptions vary
• Does not handle calendar calculations for court deadlines (separate tool)

Version History

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