Contract analysis

I. Executive Summary This memorandum analyzes the intellectual property ownership framework established under the executed Data License Agreement, referenced as AHT-PCA-DLA-2024-001, with an Effective Date of July 1, 2024 (the "Agreement"), between Pinnacle Claims Analytics LLC ("PCA"), as Licensor, and Arbor Health Technologies Inc. ("AHT"), as Licensee. The purpose of this analysis is to provide AHT's executive lea

II. Background and Transaction Context A. Parties and Their Roles Arbor Health Technologies Inc. ("AHT"). AHT is a corporation organized under the laws of the State of Delaware, with its principal executive offices located in Boston, Massachusetts. AHT is a healthcare analytics and software-as-a-service ("SaaS") company focused on predictive analytics, population health management, risk stratification, and clinical-o

III. IP Ownership Framework Under the Agreement A. PCA's Retained Ownership of Licensed Data The Agreement establishes a clear demarcation between the licensed data — which is and remains PCA's property — and the derivative works that AHT creates using the licensed data — which are and become AHT's property. We address PCA's retained ownership first.

Section 4.1 of the Agreement (Ownership of Licensed Data) provides, in relevant part: "Licensor retains all right, title, and interest in and to the Licensed Data, including without limitation all intellectual property rights therein. Nothing in this Agreement shall be construed as a sale, assignment, or transfer of ownership of the Licensed Data or any portion thereof to Licensee. The License Fee is consideration fo

1. Model Memorization Testing. Models should be tested to ensure they do not memorize or regurgitate individual claim-line records. Standard techniques include membership inference testing (assessing whether a model can distinguish between records that were and were not in the training set with unreasonable accuracy) and output inspection (verifying that model predictions do not reproduce specific claim-line field co

2. Regularization and Generalization. Standard model-training practices — including L1/L2 regularization, dropout, early stopping, and other generalization-promoting techniques — should be employed to prevent overfitting that could lead to memorization of individual records.

3. Differential Privacy Techniques. Where feasible and appropriate, AHT should consider incorporating differential privacy mechanisms into its model-training pipelines. Differential privacy provides formal mathematical guarantees that the output of an analysis does not reveal information about any individual record in the dataset. While the Agreement does not require differential privacy, its use would provide an add

4. Aggregation Thresholds. For statistical outputs, benchmarks, and reports, AHT should implement minimum cell-size thresholds (e.g., no output cell based on fewer than eleven individuals) to prevent small-cell outputs that could, in combination with external information, enable identification of individual records.

5. Derivative-Works Review. AHT's Chief Data Officer, Marcus Ellison, should implement internal policies requiring a derivative-works compliance review before any new model, analytical output, or product feature incorporating derivative works from the licensed data is distributed externally to AHT's customers or otherwise made available outside AHT's internal analytical environment. Relationship to De-Identification.

IV. Residual-Knowledge Clause A. Scope of the Clause

Section 6.4 of the Agreement (Residual Knowledge) provides: "Notwithstanding any other provision of this Agreement, Licensee's personnel may retain in unaided memory general knowledge, experience, ideas, concepts, know-how, and techniques gained through the authorized use of the Licensed Data in the ordinary course of their work for Licensee. For the avoidance of doubt, this Section does not permit Licensee's personn

1. Written Policy Statement. AHT should adopt a written policy, approved by the Chief Data Officer and General Counsel, that specifically addresses the residual-knowledge clause and the non-re-identification covenant. The policy should explain in plain language what residual knowledge AHT personnel are permitted to retain, what they are prohibited from doing, and the consequences of non-compliance (including the unca

2. Mandatory Training. All AHT personnel who are granted access to the licensed data — including data scientists, data engineers, analysts, product managers, and any contractors or consultants — must complete mandatory training on the terms of the data-license agreement, with particular emphasis on the non-re-identification covenant. Training should be completed before initial access is granted and should be refreshe

3. Exit-Interview Protocols. When an employee or contractor who had access to the licensed data departs AHT (whether voluntarily or involuntarily), AHT should conduct an exit interview that includes a reminder of the residual-knowledge clause's scope, the continuing obligation not to attempt re-identification, and the continuing confidentiality obligations under the Agreement. The departing individual should be asked

4. Access Termination. Upon departure, all access to the licensed data environment must be immediately revoked, and any devices or media containing copies of the licensed data must be returned or securely wiped, with certification provided to AHT's IT security team.

V. Negotiation History and Key Points Achieved A. PCA's Initial Position (June 12, 2024 Markup by Hargrove, Delacroix & Stein LLP) C&H circulated its initial draft of the Data License Agreement to HD&S on June 4, 2024. PCA's markup was returned by HD&S on June 12, 2024. The PCA markup contained several IP provisions that, if accepted, would have materially diminished AHT's rights in its derivative works and impaired

1. License-Back of Derivative Works. PCA initially proposed a broad, perpetual, irrevocable, worldwide, royalty-free, fully paid-up license-back to PCA for all derivative works created by AHT using the licensed data. Under PCA's proposed Section 4.2(c), PCA would have received the right to use, copy, modify, distribute, and create further derivative works from AHT's derivative works for any purpose, including PCA's o

2. Narrow Definition of Derivative Works. PCA sought to exclude trained machine-learning model weights and parameters from the definition of AHT-owned derivative works. PCA's proposed definition limited "Derivative Works" to "reports, visualizations, and aggregated statistical outputs" and expressly excluded "trained model weights, parameters, embeddings, or any other component of a machine-learning system that refle

3. Joint Ownership of Aggregated Outputs. PCA proposed that any aggregated statistical output — including benchmarks, indices, population health metrics, and actuarial tables — that was "based predominantly on the Licensed Data" would be jointly owned by PCA and AHT. The "based predominantly" qualifier was undefined, and PCA did not propose any mechanism for determining which outputs met the threshold. Under applicab

4. Elimination of the Residual-Knowledge Clause. PCA's markup deleted the residual-knowledge clause in its entirety. PCA proposed instead that "all knowledge, experience, concepts, and techniques gained by Licensee's personnel through use of the Licensed Data shall constitute Confidential Information of Licensor and shall be subject to the confidentiality obligations set forth in Section 7 of this Agreement for the d

5. Broader Non-Reconstruction Standard. PCA proposed a substantially broader non-reconstruction standard. PCA's proposed Section 4.2(b) provided that derivative works must not "reflect or be derived from the Licensed Data in a manner that permits any inference about individual claim-line-level detail." This formulation was problematic for two reasons. First, the phrase "any inference" set an impossibly low threshold:

1. License-Back Eliminated. C&H successfully rejected the license-back provision in its entirety. AHT retains sole and exclusive ownership of all derivative works, with no obligation to share, license, or provide access to them to PCA for any purpose. The final Agreement contains no license-back clause of any kind. C&H's principal argument was that AHT's derivative works are the core commercial output of AHT's busine

2. Broad Derivative-Works Definition Retained. The final definition of "Derivative Works" includes trained model weights and parameters, algorithms, analytics, and all forms of derivative insights. C&H argued that trained models are original works of authorship created by AHT's engineering efforts — reflecting AHT's proprietary model architectures, feature engineering, hyperparameter tuning, and training methodologie

3. Joint Ownership Rejected. The Agreement confirms AHT's sole ownership of all aggregated statistical outputs, including benchmarks, indices, population health metrics, and actuarial tables. C&H argued that joint ownership would create intractable co-exploitation issues, as each co-owner would independently have the right to use, license, and distribute the jointly owned outputs without accounting to the other. This

4. Residual-Knowledge Clause Preserved. C&H reinstated the residual-knowledge clause with the "unaided memory" standard. To address PCA's concerns about the potential for residual knowledge to be used to re-constitute elements of the licensed data, C&H proposed the express subordination of the residual-knowledge clause to the non-re-identification covenant — language providing that residual knowledge may be retained

5. Non-Reconstruction Standard Narrowed to Claim-Line Level. C&H negotiated the non-reconstruction standard down from PCA's proposed "any inference about individual claim-line-level detail" to the final formulation: derivative works must not "permit reconstruction of the underlying Licensed Data at the individual-claim-line level." This is a meaningfully narrower standard. The change from "any inference" to "reconstr

VI. Interaction with Other Agreement Provisions The IP ownership framework set forth in Sections III through V above does not operate in isolation. It interacts with several other key provisions of the Agreement that affect how AHT may use the licensed data, the risk profile of AHT's derivative-works activities, and AHT's obligations upon termination. We address these interactions below. A. Permitted Uses and Use Res

Section 9.3 of the Agreement grants PCA the right to audit AHT's compliance with the data-use restrictions set forth in the Agreement, including the non-reconstruction standard. Key terms of the audit provision include: (a) Frequency. PCA may exercise the audit right no more than once per calendar year. (b) Notice. PCA must provide AHT with at least thirty (30) days' prior written notice of its intent to conduct an a

Section 10.3 of the Agreement addresses AHT's obligations upon termination or expiration: (a) Return or Destruction of Licensed Data. Within sixty (60) days following termination or expiration, AHT must return to PCA or, at PCA's election, destroy all copies of the licensed data, including all raw data files, filtered extracts, subsets, and copies containing individual-level claim-line records. AHT must provide PCA w

VII. Practical Recommendations Based on our analysis of the Agreement's IP ownership framework, we recommend that AHT implement the following practical measures to ensure ongoing compliance and audit readiness. These recommendations are directed to Marcus Ellison, as Chief Data Officer, and to Helen Tsao, as General Counsel, for implementation within AHT's data-governance and legal-compliance programs. A. Internal Go