Verifiable Document Auditor
A document analysis tool that reliably counts, deduplicates, and aggregates across huge contracts, SEC filings, and discovery dumps — and shows its work with cite-back to source.
independent accountants, paralegals, and compliance consultants handling large multi-document sets
- Exact-count and distinct-count queries over million-page PDFs with cite-back to the contributing records
- 'Show your work' panel that lists every record included in an aggregate so the user can audit it
- Cross-document set operations like union, intersection, and difference on extracted entities (parties, dates, dollar amounts)
- Exportable audit trail and client-ready memo summarizing each query and its evidence
Long-context LLMs routinely hallucinate counts and aggregations beyond 100k tokens, which is the exact workload lawyers and accountants face; the paper's mergeable sketch-state method closes the gap to roughly 99% accuracy, making reliable consumer audit tools viable for the first time
Active legal/contract document-analysis market with many established vendors and roundups; Stanford/ACL papers confirm long-context hallucination is a recognized open problem that the tool would address.Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models ↗Towards Long Context Hallucination Detection ↗Top 10 Best Legal Document Analysis Software | 2026 Edition ↗
Crowded field — Relativity, Everlaw, Ironclad, LegalValidate.ai, OpusLaw, Hyperstart, and ~77 eDiscovery products already serve this exact workflow; a verifiable-aggregation angle is novel but unproven vs. incumbents.eDiscovery Software Reviews, Features & Software Guide ↗AI Contract Review & Legal Document Analysis Platform | LegalValidate.ai ↗10 Best eDiscovery Software for 2026 ↗
Proven willingness to pay — Relativity publishes pay-as-you-go and annual eDiscovery pricing, and ComplexDiscovery tracks 24 eDiscovery offerings; accountants/paralegals/compliance consultants have established software budgets.eDiscovery Pricing Overview - Winter 2024 - ComplexDiscovery ↗Legal Data Software Pricing - Relativity ↗
Compliance, audit, and eDiscovery workflows are regulation-driven and durable; long-context hallucination is documented as an open problem (ACL 2025, Stanford JLA), so verification tooling stays relevant as LLMs are adopted.The rise of hallucination in large language models ↗Ultimate guide to contract analytics for legal professionals ↗
Buildable — many competitors ship comparable pipelines, but the claimed 99% aggregation accuracy with cite-back is non-trivial; the cited arXiv paper could not be located in search to validate the technique, raising execution risk.PDF Enhancing Legal Document Analysis with Large Language Models ↗