Does It Tie Out? Towards Autonomous Legal Agents in Venture Capital

TL;DR

The paper tackles capitalization tie-out in venture-financing datarooms, a high-stakes verification task where LLM-driven agentic methods struggle with multi-document reasoning and traceability. It contrasts a lazy RAG-based approach with an eager world-model architecture (Equall) that builds a layered representation of the company’s lifecycle (Foundational Extraction, Inductive Event Modeling, and Neuro-Symbolic Verification) to produce a verifiable virtual cap table. Empirical results show non-linear growth of evidentiary burden and a shift in anomaly types as governance matures, with Equall achieving substantially higher F1 scores and faster verification per check than baselines. The authors argue that explicit world-model construction (via an Event Graph) provides a generalizable foundation for robust, autonomous legal reasoning and broader applied legal intelligence.

Abstract

Before closing venture capital financing rounds, lawyers conduct diligence that includes tying out the capitalization table: verifying that every security (for example, shares, options, warrants) and issuance term (for example, vesting schedules, acceleration triggers, transfer restrictions) is supported by large sets of underlying legal documentation. While LLMs continue to improve on legal benchmarks, specialized legal workflows, such as capitalization tie-out, remain out of reach even for strong agentic systems. The task requires multi-document reasoning, strict evidence traceability, and deterministic outputs that current approaches fail to reliably deliver. We characterize capitalization tie-out as an instance of a real-world benchmark for legal AI, analyze and compare the performance of existing agentic systems, and propose a world model architecture toward tie-out automation-and more broadly as a foundation for applied legal intelligence.

Figures (11)

• Figure 1: The tie-out workflow. Lawyers review and cross-reference heterogeneous dataroom documents—one or more capitalization tables, and supporting legal documentation (e.g., SAFEs, option agreements, amendments, cancellations, etc.)—to understand the legal reality of the company and ensure the capitalization table is accurate by comparing it against the non-cap table documents, which provide the ground truth. The output: verified legal positions with document traceability, and flags for discrepancies requiring further review and/or action.
• Figure 2: Document category. Red dashed lines indicate Zipf's law fits ($R^2 \sim 1$ is perfect fit).
• Figure 3: Real-world example of a grant lifecycle. A single stock option grant that is later repriced, then affected by a 10:1 stock split, then partially exercised into common stock (with the remainder expiring), after which the resulting common shares are sold or transferred to three different holders.
• Figure 4: Comparison of key dataroom statistics, including total pages, documents, securities, and shareholders, across companies in different financing stages.
• Figure 5: Distribution of categories across four companies at varying financing stages.