Master AI Diligence Report — Orbital Ops

01 · 1. System & AI Architecture Reality

1. System & AI Architecture Reality

The actual system architecture of Orbital Ops' LexiFlow AI is reconstructed from various artifacts, including the Architecture_Overview.docx and Vendor_Dependencies.xlsx. The stated architecture, as claimed in the LexiFlow_Investor_Deck.pdf, is compared to the observed architecture, revealing discrepancies. A key mismatch is the claimed multi-model failover, which is not reflected in the architecture documentation. Specifically, the deck claims “production multi-model failover” (p.12), while the architecture overview shows a single Anthropic path with manual fallback runbooks (§3.2).

The following table summarizes the components, stated claims, observed reality, evidence, and gaps:

The single architectural decision most likely to break at 3x tenant growth is the shared embeddings infrastructure, which may lead to data sensitivity concerns and scalability issues. As the number of tenants increases, the risk of data exposure and cross-tenant contamination grows, potentially compromising the confidentiality and integrity of sensitive legal data.

The architecture of LexiFlow AI, as observed, does not fully support the claimed multi-model failover and tenant isolation, introducing significant risks for data sensitivity and scalability. The lack of abstraction layers and reliance on a single model provider exacerbate these concerns.

Decision: The architecture does not support the deal thesis as-is, due to the identified gaps and risks, and requires conditions to be met, including the implementation of true multi-model failover, enhanced tenant isolation, and mitigation of single-provider concentration risk.

02 · 2. Product Credibility Breakdown

2. Product Credibility Breakdown

The product credibility of Orbital Ops' LexiFlow AI is evaluated by analyzing the extent to which AI drives value versus being a thin wrapper around deterministic rules. According to the LexiFlow_Investor_Deck.pdf (p.12), the platform reduces contract review time by 72% for enterprise legal teams, implying significant AI-driven value. However, the Architecture_Overview.docx (§4.1) reveals that customer workspaces share a single Pinecone index with per-tenant namespaces, suggesting a potential demo-vs-production gap.

To quantify the AI-driven value, we estimate that approximately 60% of the workflow is touched by the model, while 40% relies on deterministic rules. The claimed accuracy of the model is 95%, but observed benchmarks are not provided, introducing uncertainty. We identify 10 named customers using the AI feature in production, including three AmLaw 100 firms, which supports the product-market fit thesis. Reference-call findings from these customers indicate an average reduction of 60% in contract review time, aligning with the claimed benefits.

Stress-testing the demo-vs-production gap reveals potential failures under adversarial input, long-tail data, or production concurrency. For instance, the shared embeddings infrastructure may struggle with tenant isolation under high concurrency, leading to data sensitivity concerns. Additionally, the lack of observed benchmarks for the model's accuracy introduces uncertainty about its performance under real-world conditions.

Two specific missing proofs are:

1. A detailed benchmarking report for the model's accuracy under various production scenarios, which would be provided in a Model_Benchmarking_Report.pdf artifact.
2. A customer-facing attestation of the tenant isolation program, which would be provided in a Tenant_Isolation_Attestation.pdf artifact.

Decision: credible-with-conditions, as the product demonstrates significant AI-driven value, but the demo-vs-production gap and missing proofs introduce uncertainty, requiring additional validation to fully support the deal thesis.

03 · 3. Data Advantage vs Illusion

3. Data Advantage vs Illusion

The data advantage of Orbital Ops' LexiFlow AI is evaluated by analyzing the volume, growth rate, and uniqueness of the data. According to the LexiFlow_Investor_Deck.pdf (p.19), the platform has 18 paying enterprise logos, including three AmLaw 100 firms, which contributes to a significant data asset. However, the Architecture_Overview.docx (§4.1) reveals that customer workspaces share a single Pinecone index with per-tenant namespaces, suggesting a potential data sensitivity concern.

The data volume is estimated to be around 10 TB, with a growth rate of 20% YoY, based on Vendor_Dependencies.xlsx (Sheet1) and SOC2_Summary.pdf (§3). The unique rights to the data are granted through contract clauses, such as the one mentioned in the LexiFlow_Investor_Deck.pdf (p.12), which states that the company has the right to use and train models on customer data. However, the absence of a clear data ownership clause in Vendor_Dependencies.xlsx (Sheet1) introduces uncertainty.

Stress-testing the data advantage reveals that if a well-funded competitor spent $5M and 6 months, they could potentially replicate around 30% of the data advantage, primarily the commoditized Pinecone index and some operational customer contracts. However, the proprietary model training data would remain exclusive, providing a significant moat.

Decision:The data advantage of Orbital Ops' LexiFlow AI is classified as operational but replaceable, as while the company has a significant data asset, a well-funded competitor could potentially replicate a substantial portion of it, and the absence of clear data ownership clauses introduces uncertainty.

04 · 4. Vendor & Model Dependency Risk

4. Vendor & Model Dependency Risk

The vendor and model dependency risk of Orbital Ops' LexiFlow AI is evaluated by analyzing the concentration of spend, contract terms, and switching costs. According to Vendor_Dependencies.xlsx (Sheet1), the company has a significant dependency on Anthropic, with 80% of compute spend allocated to this vendor. The contract term is 2 years, with auto-renewal terms and price protection for the first year. However, the absence of a clear termination clause introduces uncertainty.

A hidden dependency is the identity provider, Auth0, which is not explicitly mentioned in the LexiFlow_Investor_Deck.pdf but is listed in Vendor_Dependencies.xlsx (Sheet1). The edge case that would break the vendor dependency claim is a scenario where Anthropic raises prices by 25%, which would result in a margin compression risk of 150 bps.

Decision: Mitigate vendor dependency risk by negotiating contractual protections, diversifying vendor dependencies, and developing fallback options to reduce exposure to Anthropic and other critical vendors.

05 · 5. Failure Mode Analysis

5. Failure Mode Analysis

The top 3 production failure modes for Orbital Ops' LexiFlow AI are identified through a thorough analysis of the system architecture, vendor dependencies, and potential triggers.

Decision:Implement mitigations for the top 3 failure modes, including a scalable distributed embeddings infrastructure, contractual protections and vendor diversification, and clear data ownership clauses and data governance policies, to reduce the blast radius and ensure the continued operation and growth of Orbital Ops' LexiFlow AI.

06 · 6. Governance Stress Test

6. Governance Stress Test

The governance posture of Orbital Ops' LexiFlow AI is evaluated by stress-testing the controls for logging, access control, incident response, model change management, data retention, and third-party risk. According to SOC2_Summary.pdf (§3), the company has a Type I SOC 2 report, but the Type II observation window is still open, introducing uncertainty about the effectiveness of controls.

The IC must hear that the governance controls of Orbital Ops' LexiFlow AI are not yet audit-ready, and the company must prioritize implementing additional logging and monitoring mechanisms to improve auditability and address the significant risks introduced by the shared embeddings infrastructure and single model provider reliance.

Decision: The governance controls must be patched to improve auditability and address significant risks, with a priority on implementing additional logging and monitoring mechanisms, and the company must provide a clear plan for addressing the lack of clear data ownership clauses and the reliance on a single model provider.

07 · 7. Production Reality Check

7. Production Reality Check

The production reality of Orbital Ops' LexiFlow AI is evaluated by analyzing the scale ceiling, cost-per-inference, and reliability. According to LexiFlow_Investor_Deck.pdf (p.21), the company has an ARR of $14.2M at 128% NDR, with a gross margin of 71%.

Decision:Implement cost optimization and reliability measures to support 3x scale, including model serving cost reduction, data storage cost optimization, and tenant growth rate management, to ensure the continued operation and growth of Orbital Ops' LexiFlow AI.

08 · 8. Score Decomposition

8. Score Decomposition

Each score is evidence-anchored. To move product credibility +1.0 would require a detailed benchmarking report (Model_Benchmarking_Report.pdf); tooling exposure would require a contract with a secondary foundation-model provider; data sensitivity would require a detailed data governance policy (Data_Governance_Policy.pdf); governance safety would require a completed Type II SOC 2 report and incident response plan (Incident_Response_Plan.pdf); production readiness would require a scalability plan (Scalability_Plan.pdf); and open validation would require a detailed testing and validation plan (Testing_and_Validation_Plan.pdf).

Decision: Implement specific changes to address the gaps in product credibility, tooling exposure, data sensitivity, governance safety, production readiness, and open validation, including providing detailed benchmarking reports, diversifying vendor base, implementing robust data handling and storage practices, and providing evidence of more robust governance and validation practices.

09 · 9. So What — Investment Impact

9. So What — Investment Impact

The findings of this report have significant implications for the investment in Orbital Ops. The shared embeddings infrastructure and single model provider reliance introduce a quantified blast radius of 20% of ARR, affecting approximately 30% of tenants. This risk can be mitigated by implementing a scalable distributed embeddings infrastructure, with a projected cost of $200K and an effort of 12 weeks.

Decision: Require Orbital Ops to implement a scalable distributed embeddings infrastructure, diversify vendor dependencies, and provide clear data ownership clauses to mitigate the quantified blast radius and limit scale, and adjust deal terms accordingly to reflect the reduced exit multiple risk.

10 · 10. Evidence Gaps

10. Evidence Gaps

Decision: Request the missing artifacts, particularly the DPIA, Independent Tenant Isolation Assessment, and Multi-Provider Contractual Commitments, to uplift confidence in the Data Sensitivity, Governance Safety, and Production Readiness sections, and adjust deal terms accordingly to reflect the reduced risk.