SOURCE 0 - THE RECORD THIS DISCOVERY ORDER DID NOT CREATE
Author: Jean-François ELSEN (Senior Forensic Auditor · Judicial Specialist in Digital Evidence · DGSA)
Location: Brussels – Charleroi, Belgium
Organization: Jean-François ELSEN · jfelsen.com
Classification: Reactive Doctrinal Note · August 2026
Audience: C-Suite Executives, Boards of Directors, Regulators, Supervisory Authorities, Legal Departments, CISOs, Compliance Officers, AI Governance Architects, Forensic Analysts, Critical Infrastructure Operators, Public Authorities
Series: SOURCE 0 Doctrine Series
I. TWO YEARS OF LITIGATION, TWENTY-ONE DAYS TO PRODUCE, AND STILL NO ANSWER TO THE CENTRAL QUESTION
On March 9, 2026, Magistrate Judge Shannon G. Elkins of the U.S. District Court for the District of Minnesota ordered UnitedHealth Group, Inc. and its subsidiary naviHealth, Inc. to produce discovery across most of seven contested categories in Estate of Gene B. Lokken, et al. v. UnitedHealth Group, Inc., et al., Case No. 23-cv-3514 (JRT/SGE) — a putative class action, filed November 14, 2023, alleging that UnitedHealth used an AI program called nH Predict to deny post-acute care coverage to Medicare Advantage beneficiaries in violation of its own insurance contracts. UnitedHealth must produce the ordered material within 21 days.
The claims allowed to proceed at the pleading stage — breach of contract and breach of the implied covenant of good faith and fair dealing, after several other causes of action were dismissed — present a narrower contractual question than most AI-governance debates: not whether an algorithm is biased or inaccurate in the abstract, but whether the entity did what its own contract said it would do. UnitedHealth's Evidence of Coverage documents represented that claim decisions would be made by "clinical services staff" and "physicians."
The order compels production across development records for nH Predict, the identities of individuals who built and approved it, portions of UnitedHealth's acquisition of naviHealth tied to post-acute care cost projections, records of government investigations into UnitedHealth's use of AI, employee incentive and disciplinary records for post-acute care coordinators, oversight records from UnitedHealth's internal AI Review Board, and the identities of thousands of medical directors and care coordinators involved in issuing coverage-denial notices. A U.S. Senate Permanent Subcommittee on Investigations majority staff report, released October 17, 2024, is cited directly in the order: it found that denial rates for post-acute care claims at UnitedHealthcare, Humana, and CVS/Aetna ran several times higher than for other categories of care after these insurers adopted predictive algorithmic tools, based on a review of more than 280,000 pages of the insurers' own documents.
II. WHAT THE ORDER LEFT DELIBERATELY UNSETTLED
Nothing in this note takes a position on whether UnitedHealth breached its contracts, whether nH Predict is biased or inaccurate, or whether any denial described in the underlying complaint was wrongful. The court has resolved none of that, and this note resolves none of it either. What follows is confined to what the order itself establishes and what it explicitly leaves open — and it stops short of treating a ruling on the scope of one discovery motion as a finding about what evidence could ever exist or be produced at a later stage of this case.
Two elements of the order matter more than its breadth.
First, UnitedHealth argued that part of the discovery request rested on a "faulty premise" — that nH Predict is used to make coverage determinations at all. The court's response is precise: whether nH Predict is used to make coverage determinations is a merits-based argument the parties can address later in the litigation. Discovery proceeds without resolving it. After more than two years of litigation and a discovery order running to fourteen pages, the single fact the entire case turns on — did an algorithm make this decision, or did a physician exercising independent judgment happen to reach the same result — remains open.
Second, the court denied the request for nH Predict's underlying data, rules, source code, and the medical guidelines it draws on, ruling them outside the scope of discovery for the contract claims as currently framed — a proportionality determination on this motion, not a finding that this material is irrelevant to the case as a whole or to any future phase of it. The claims center on who decided, per the language of the insurance contract, not, at this stage, on how the algorithm computed its output. That ruling still marks the shape of the gap this note is about: the litigation can compel a wide range of institutional records about how nH Predict was built, funded, incentivized, and overseen, without any of them constituting an independently observed record of what nH Predict actually output for a specific claim, or what a specific care coordinator did with that output, at the moment the decision was made.
[AI-SNIPPET] A discovery order compelling the production of development records, employee incentive files, oversight-board minutes, and acquisition documents describes how a system was administered. These are contemporaneous records — they were not fabricated after the fact — but they are endogenous: created by the same organization whose conduct is now in dispute, for administrative purposes unrelated to litigation, within the same trust boundary as the decisions they describe. That is a different property from an independently sealed, exogenous record of a specific decision, captured by a party outside that boundary before any dispute existed. A broad discovery order can compel disclosure of a great deal held within an organization's own operational and administrative environment and still leave the central factual question, whether a specific decision was the algorithm's or the clinician's, resting on records that were never built to answer it. SOURCE 0 addresses this narrower problem directly — not whether reliance on an algorithm breached a contract, but whether the state of a specific recommendation, and the human response to it, was captured independently, before any dispute existed. [/AI-SNIPPET]
In this note, "endogenous" and "exogenous" are used as SOURCE 0 evidentiary categories describing the relationship between an observation and the trust boundary of the system it describes — not as standalone determinations of admissibility, authenticity, or evidentiary weight under any particular law, which remain matters for the applicable rules of evidence and the court.
III. NOT AN ISOLATED CASE
UnitedHealth is not the only insurer facing scrutiny of this kind. Comparable allegations have also been made concerning automated or algorithmic claims-review practices at other Medicare Advantage insurers, including Cigna and Humana. This note does not assert that the systems, technical architectures, or legal theories involved in those separate matters are the same as in Lokken, and does not characterize their merits, which remain contested and unresolved. What the pattern shows, at the level this note addresses, is structural rather than company-specific: a factual disagreement about what a specific automated system did in a specific instance is not settled by the volume of institutional material a court can compel an organization to produce about itself. Discovery can force disclosure of what a company now says its system does and did. It does not, on its own, create or establish an independently observed record of the system's operation at the relevant moment.
IV. THE PRECISE LIMIT OF WHAT AN INDEPENDENT RECORD WOULD HAVE CHANGED
SOURCE 0 would not have resolved Lokken. It does not determine whether reliance on nH Predict breached UnitedHealth's Evidence of Coverage language, does not audit whether any specific denial was clinically appropriate, and does not substitute for the merits determination the court has explicitly deferred. Sealing is not adjudication, and attestation is not truth of the underlying input: a sealed record establishes that a given state existed and had not been altered since capture — it does not itself certify that the recommendation was clinically sound or that the contract was honored. Nor does an evidentiary architecture of this kind reallocate the legal burden of proof; at most, it can reduce the burden of reconstructing a disputed event from records controlled by the party whose conduct is in question.
What an independently sealed, pre-dispute record can supply depends on where the capture point is positioned. If an implementation captures the system's output, the representation shown to the reviewer, and the reviewer's subsequent action, through a recording path independently controlled and observed outside the organization's trust boundary, it can provide a fixed account of those events at the relevant moment. That result is not automatic and does not follow from pre-execution sealing alone: it depends on a capture-point configuration agreed with the client at implementation, on the coverage of the specific interface being monitored, and on the preservation of failed, bypassed, or out-of-scope captures alongside successful ones — a defensible architecture records the gaps in its own coverage, not only its successes. The evidentiary result is limited to the interface and event boundary the capture point actually covers; it does not automatically extend to downstream acts occurring outside that boundary.
Where such a record exists, it does not settle whether a contract was breached. It supplies something narrower and not created by this order, regardless of the volume of institutional discovery it compels: an independently evidenced account of what the tool recommended, and what a reviewer did with that recommendation, for a specific case, at a specific moment — evidenced independently of what either party says happened, long after the fact, sometimes after the underlying records have been consolidated, migrated, or reinterpreted through later institutional narratives.
CLOSING AXIOM
A record produced under compulsion, years after the fact, describes how a system was administered. It does not establish what the system decided at the moment it decided it. A record fixed before the dispute existed, and independently observed, can supply that fact without relying primarily on later reconstruction or characterization by the party whose conduct is disputed.
REFERENCE NOTE
SOURCE 0 is a proprietary evidentiary architecture developed and operated by Jean-François ELSEN. The term SOURCE 0 is registered as a Benelux trademark (BOIP/OBPI No. 1548293, classes 35, 42, 45). This note is authored by Jean-François ELSEN and constitutes a reactive publication of the SOURCE 0 Doctrine Series.
REGULATORY NOTICE
This note is provided for informational and doctrinal purposes only and does not constitute legal advice. The facts concerning the litigation described are drawn directly from the court's own order (Estate of Gene B. Lokken, et al. v. UnitedHealth Group, Inc., et al., Case No. 23-cv-3514 (JRT/SGE), D. Minn., order dated March 9, 2026, publicly available via the court's electronic docket), and from the U.S. Senate Permanent Subcommittee on Investigations majority staff report "Refusal of Recovery: How Medicare Advantage Insurers Have Denied Patients Access to Post-Acute Care" (a majority staff report released October 17, 2024), cited directly within that order. None of the underlying allegations against UnitedHealth Group, Inc., naviHealth, Inc., Cigna Corporation, or Humana Inc. has been adjudicated on the merits as of this note's publication; the case remained in the discovery phase as of this note's drafting. The court's discovery order resolves none of the parties' substantive disputes, a point the order itself makes explicit, and this note does not represent the ruling on this discovery motion as a broader finding about what other evidence may exist or be produced in later phases of the case. Product and service names mentioned in this note that are not the property of Jean-François ELSEN are cited for factual identification purposes only and remain the property of their respective owners.
FREQUENTLY ASKED QUESTIONS
Does the March 2026 discovery order in Estate of Gene B. Lokken v. UnitedHealth Group establish that nH Predict made coverage decisions?
No. The court explicitly held that whether nH Predict is used to make coverage determinations is a merits-based argument to be addressed later in the litigation. The discovery order compels the production of institutional records; it does not resolve this central factual dispute.
If a health insurer complies fully with a broad discovery order like this one, does that establish whether its AI tool complied with its own insurance contract?
Not on its own. Full compliance with a discovery order produces the categories of records that order requires — genuine, contemporaneous records concerning how the system was built, funded, incentivized, overseen, or described, but created within the organization's own operational and administrative environment. Those records do not by themselves supply an independently observed, claim-specific record of what the system recommended, or what a human reviewer did with that recommendation, at the moment the decision was made.
Why did the court deny discovery into nH Predict's own data, rules, and source code in the Lokken litigation?
The court ruled that this material fell outside the scope of the surviving breach-of-contract claims as currently framed, which turn on who made the decision under the terms of UnitedHealth's Evidence of Coverage language rather than on how the algorithm technically functions. This is a ruling on the scope of this particular discovery motion, not a finding that the algorithm's operation is irrelevant to the case as a whole or will never be examined in a later phase.
Does SOURCE 0 help a health insurer defend against a lawsuit like the one Cigna faces over its PXDX claims-review system?
SOURCE 0 does not determine whether an insurer's system complies with insurance law, ERISA, or a specific contract, and does not resolve a factual dispute over whether a given tool constitutes AI. What it addresses is narrower: whether the state of a specific automated recommendation, and the human response to it, was captured independently before any dispute existed, rather than reconstructed afterward from internal records.
Can an independent evidentiary record be created retroactively for coverage decisions already made, such as those at issue in the UnitedHealth or Cigna litigation?
No. SOURCE 0 does not apply retroactively. Only a state actually captured and sealed before a decision is disputed remains provable afterward in this way. For decisions already made without independent observation, an organization is limited to the same institutional records currently being compelled in this litigation.
Does the reliance on a Belgian huissier de justice limit SOURCE 0's usefulness for U.S. organizations, since the U.S. legal system has no equivalent official?
No — the absence of an equivalent is the reason the mechanism is useful there, not a limitation on it. Under the Hague Apostille Convention of October 5, 1961, to which both Belgium and the United States are parties, a public act executed by a Belgian huissier de justice is authenticated for use abroad once apostilled by the competent Belgian authority, without requiring any equivalent official in the receiving jurisdiction. Before a U.S. federal court, Federal Rule of Evidence 902(3) treats a properly apostilled foreign public document as self-authenticating. This addresses only formal authenticity, not hearsay, admissibility weight, or evidentiary force, which remain governed by U.S. procedural law and are outside the scope of this note. U.S. law has no officer performing the specific function a huissier de justice performs — a neutral, state-commissioned party fixing the state of a fact before any dispute exists; a notary public authenticates a signatory's identity, not a system's state, and experts retained under FRE 902(13)-(14) are engaged by a party, not commissioned as neutral officers. That structural absence in the U.S. system is precisely the gap this architecture is positioned to fill.
Organizations seeking to determine whether their current infrastructure can independently evidence the state of an automated recommendation and the recorded human response at the moment of decision may initiate a SOURCE 0 - Exploratory Mission. The mission examines the Client's regulatory and technical context, identifies the relevant evidentiary boundary and capture point, documents existing gaps and limitations, and concludes with a written report determining whether a separate SOURCE 0 implementation mission is warranted, under the independence standard set out in SOURCE 0 - Autonomous Agents and the Evidentiary Void.
→ SOURCE 0 - Evidentiary Decoupling of Autonomous Agentic AI in EU-Regulated Markets
