Type-Checked Compliance: Deterministic Guardrails for Agentic Financial Systems Using Lean 4 Theorem Proving

This is a 4-day-old arxiv paper with zero signals of adoption, implementation, or community traction. The core insight—using formal verification (Lean 4) instead of probabilistic classifiers for financial compliance—is a novel combination of existing techniques (formal methods + LLM guardrails), but the project exists only as a paper, not a working system. Zero stars, forks, and velocity confirm this is pre-prototype. DEFENSIBILITY (3): The paper articulates a real problem (probabilistic guardrails are insufficient for financial compliance) and proposes a conceptually sound solution (formal verification). However, there is no evidence of: (1) working implementation, (2) user adoption, (3) ecosystem, or (4) moat beyond the idea itself. Once published, the approach is reproducible by any team with Lean 4 expertise. The technical barrier is high but not insurmountable for frontier labs or well-resourced teams. FRONTIER RISK (high): Frontier labs (OpenAI, Anthropic, Google) are actively building safety and compliance infrastructure for AI agents. Formal verification is table-stakes in financial AI safety. Anthropic's Constitutional AI, OpenAI's fine-tuning for compliance, and Google's responsible AI initiatives all touch this space. A frontier lab could easily: (a) hire Lean experts, (b) build similar formal verification layers into their agent frameworks, or (c) acquire this capability. The compliance-specific domain narrows the audience but increases attractiveness to financial-focused API providers (e.g., JPMorgan's COiN, institutional AI platforms). NOVELTY (novel_combination): Formal verification is not new. LLM guardrails are not new. The application of theorem proving to financial compliance guardrails is a meaningful new pairing, but it's an engineering challenge rather than a breakthrough in either formal methods or finance. The paper likely contributes a methodology and proof-of-concept, not a fundamentally new algorithm. IMPLEMENTATION DEPTH (theoretical): Described as a paper. No code repository, no working system, no deployment evidence. This is purely a reference implementation or position paper. COMPOSABILITY (theoretical): As a paper, the primary contribution is the architectural framework and proof methodology, not reusable code. Integration would require teams to independently implement Lean 4 proofs for their financial agents. Not a pip-installable library or API. CRITICAL RISK: The project will be obsolete if: (1) Frontier labs publish competing formal verification frameworks for LLM guardrails, (2) Industry adopts lighter-weight verification approaches that achieve 95% of the compliance benefit at 20% of the formalization cost, or (3) Regulatory bodies accept probabilistic guardrails + audit trails as sufficient (most likely given existing precedent in automated trading). The window to convert this paper into a real product with financial customer lock-in is narrow (12-18 months before frontier labs or financial incumbents build similar capabilities in-house).