RESEARCH DOSSIER
Claims Denial Reason Codes and Explainability Quality
Abstract. A claims denial is an adverse determination concerning an asserted insurance benefit, a payment component, or an entitlement to service. Reason codes compress the grounds of that determination into operational categories. Explainability quality concerns the extent to which a decision can be reconstructed from applicable terms, material evidence, adjudicative reasoning and the available route to correction. This monograph studies these objects through comparative legal and supervisory analysis, health-payment data architecture, selected administrative investigations, explainable artificial intelligence literature and explicitly illustrative quantitative methods. It establishes distinctions between disposition, contractual ground, evidence deficiency, workflow state, payment adjustment and appeal outcome; develops practical criteria for documenting each ground in full and partial denials; and examines the effects of code granularity, aggregate reporting, automation, institutional incentives and language on accountability. The principal analytical finding is that a faithful code and a comprehensible explanation are complementary forms of access to the same adjudication record. Neither alone establishes evidentiary sufficiency or contractual correctness. A defensible process keeps the operative decision, supporting evidence, policy version, human responsibility and corrective mechanism mutually traceable, including when a predictive system contributes to a decision. The work presents original comparative synthesis and reproducible hypothetical calculations, without claiming a new theory, proprietary data analysis, causal estimates of industry behavior or universally applicable legal rules.
Keywords: insurance claims; claims denial; adverse benefit determination; reason codes; partial denial; claims adjudication; contractual grounds; explanation fidelity; remittance advice; procedural fairness; appeals; insurance governance; consumer protection; audit trail; explainable artificial intelligence; disclosure; inter-rater reliability; claims analytics.
Introduction
The adjudication of insurance claims transforms a report of loss, a request for benefits or an asserted obligation into an institutional determination. That determination can authorize payment, decline coverage, recognize only part of a demand, request additional evidence, or defer decision under an applicable procedure. Each result carries consequences for different parties: the policyholder or beneficiary, the treating provider or repairer, the insurer, a reinsurer where relevant, and the agency responsible for oversight. The decision also becomes a record from which an organization estimates expenses, improves operations and learns about defects in contractual language.
The scientific object of this dossier is the reason attached to an adverse determination. The related fields of claims science, market conduct and consumer protection contribute distinct units of analysis. Claims science examines the adjudication itself; market conduct examines the organizational arrangements that make adjudication reliable; consumer protection examines intelligibility and effective access to correction. Their conjunction explains why a technically valid message can still be inadequate for its intended recipient.
Three questions organize the study. What grounds must a reason code distinguish in order to preserve the nature of an adverse determination? How can case-specific explanations be evaluated for accuracy, clarity and practical usefulness? How can an organization preserve accountability when algorithms, rules engines, outsourced services or generative systems participate in producing a denial? The questions apply to multiple insurance lines while recognizing that a medical service authorization, a property indemnity decision and a liability defense determination have different contractual and procedural structures.
The research design is comparative documentary analysis, supplemented by structured measurement proposals and hypothetical examples. The primary source set comprises identifiable regulatory texts, supervisory instruments and agency documentation; selected published empirical reports and explanation research provide methodological and observational context. Sources are examined at the level relevant to the statements made. No sampling frame of insurer claim files, experiment with real claimants, full legal survey or independent replication of government microdata is represented as having been conducted. The reference date is November 19, 2026; where a dated legal compilation or original enactment is used, its temporal scope remains explicit.
A denial is not inherently evidence of misconduct. Insurance protects a specified risk subject to terms, limits and applicable law, and an appropriately established adverse determination can be a necessary part of contract performance. The scientific question is whether the record establishes that this determination follows from the applicable basis and whether the affected party can understand and challenge it through the relevant mechanism. A low denial rate does not prove an accurate process, any more than a high denial rate establishes wrongdoing without examination of the underlying obligations and population.
The claim determination as an information object
A claim may concern one event, multiple exposures, several insured interests, numerous payment lines, multiple persons or a sequence of requested benefits. A determination is therefore best analyzed at a declared unit: for example, claim, coverage component, service line, authorization request or distinct benefit period. The same event may legitimately generate several determinations. Counting every adjustment code as a denied claim confuses messages with adjudicative units.
Disposition records what happened to a request: accepted, partly accepted, denied, pending, withdrawn, superseded or another defined result. A denial ground identifies the factual and normative basis of an adverse decision. A workflow status identifies where a request is in transaction processing. Evidence status records which materials are present, absent, disputed or under review. A payment adjustment records a difference between submitted and recognized financial amounts. A review outcome records a later determination in the appeal or reconsideration process. These variables sometimes coexist in a single legacy field, but their semantic distinction is necessary for reliable analysis.
Consider a hypothetical property claim for $30,000 consisting of $18,000 for covered building damage, $8,000 for excluded wear-related deterioration and $4,000 for expenses that remain under investigation. Assume a valid $1,000 deductible applies to the covered loss and no other adjustment. The current undisputed payment is $17,000; $8,000 is adversely determined on the stated exclusion; $4,000 is pending. The deductible is a contractual sharing of covered loss, not by itself an adverse coverage determination. If the system labels the entire $13,000 difference as denial, it merges three economically and legally different quantities. If it labels the whole claim closed and paid, it suppresses the pending component.
In a health-payment transaction, a provider may submit a charge exceeding a contractual fee schedule. A reduced allowed amount can be an ordinary contractual adjustment rather than a refusal to recognize the patient's entitlement to coverage. A duplicate submission can be rejected administratively even though the original submission was payable. An accurate operational vocabulary must preserve these differences before any performance statistic is calculated.
The distinction also extends to finality. A provisional adverse position subject to further investigation, a first-level denial, a denial following internal review and a final external adjudication are not interchangeable. Each belongs to a dated decision event. A corrected denial must remain traceable without continuing to appear as the current final result. An event history, rather than an overwritten latest-status field, provides the information needed to analyze error correction and delay.
Historical and institutional foundations
Claims explanations arose from the contractual and evidentiary demands of insurance long before digital classification. A decision had to communicate which promise was implicated, whether its conditions were satisfied and what consequence followed. Industrial scale, specialized claims administration and electronic transactions created incentives to standardize recurring reasons. Standardization reduced the cost of routing a file, identifying required evidence and reconciling payments. It also made the explanation vulnerable to losing the circumstances that originally justified the classification.
The institutional context is represented by the IAIS claims-handling standard, ICP 19.10. Its supervisory approach connects fair and timely claims handling to the information supplied to customers, including reasons for full and partial denials and how a dispute can be pursued. The guidance also retains insurer responsibility for claims functions performed through outsourcing. The standard provides a comparative benchmark for insurance supervision; its particular implementation depends on jurisdiction.
The NAIC's (1997) Unfair Claims Settlement Practices Act illustrates an earlier state-oriented model of claims conduct. Its listed practices link adequate investigation to accurate representations of coverage and reasonable explanations of denials or compromise offers. It is a model statute and cannot establish the law of a particular U.S. state without examination of that state's enactment. In the United Kingdom, the FCA's ICOBS 8.1 claims rules provide a distinct set of operational conduct requirements, including fair and prompt handling, guidance and avoidance of unreasonable rejection.
These instruments express a common institutional concern: the claims organization must make a decision it can substantiate and provide information appropriate to the affected party. Their legal form differs. International supervisory standards influence supervisory design, model laws propose statutory language, and national rules operate within their own legal fields. A comparative claims code architecture may embody shared informational requirements while keeping the legal authority governing each determination separately identifiable.
Digital remittance offers a particularly clear illustration of how a standardized code can support operations without substituting for the decision record. CMS describes the electronic remittance advice used in Medicare billing, including standardized claim adjustment reasons, remark codes and provider-level adjustments. These codes support financial reconciliation and workflow at scale. They cannot be treated as a universal classification of all insurance denials or assumed to supply every claimant-facing disclosure required in another legal setting.
A semantic architecture for denial grounds
A reason taxonomy is a controlled vocabulary for classifying substantively meaningful causes of an adverse decision. Its design should start with adjudication and contractual coverage rather than the fields that happen to exist in a software implementation. The following categories are practical analytical groupings derived from recurring insurance decision problems. They are illustrative and do not purport to replace official remittance code sets or enact a new doctrine.
Eligibility and attachment grounds concern whether the relevant insurance relationship, covered person, property, risk, benefit period or insured interest falls within the applicable contract. Examples may include absence of applicable coverage at the relevant time or ineligibility under a plan. The record must identify the relevant dates or classification and any evidence that affects them. A billing enrollment mismatch may require correction rather than support a final coverage refusal.
Insured-event and triggering-condition grounds concern whether the contractually defined event or threshold has been established. A benefit may depend on death, accidental injury, inability to perform defined duties, physical loss, liability or another specified circumstance. Distinguish absence of the event from inability to establish it on currently available evidence.
Scope and exclusion grounds arise when an otherwise asserted loss falls outside the coverage promise or within a particular exclusion. A valid explanation identifies the operative policy provision, relevant exception, material facts and relationship between them. Exclusions must not be silently merged with conditions, definitions or deductibles because these may carry different interpretive and evidentiary implications.
Causation grounds concern the relationship between an event, mechanism of loss and the insured consequence. The same physical damage can result from several contributing mechanisms. The code should capture the adjudicated issue without pretending that a broad label establishes how causation was determined. Expert evidence, alternative explanations, policy language and relevant law require separate examination.
Limit, aggregation and valuation grounds concern amounts rather than necessarily the existence of coverage. Exhaustion of a specified benefit, application of an applicable sublimit, allocation among covered and uncovered items, and disagreement about valuation all may produce partial adversity. Contractual cost sharing, ordinary fee schedules and agreed pricing adjustments should retain distinct classifications. The amount subject to denial must be reconciled to the component identified by the code.
Procedural and evidence-related grounds concern requirements for notice, proof, documentation, authorization or timely submission, insofar as those requirements apply. Distinguish information currently missing from documents already supplied but judged insufficient. Distinguish both from an insurer's definitive conclusion that substantive conditions are not satisfied. Whether a procedural defect can justify a denial depends on applicable law and contract; the existence of an internal deadline or system rule alone is insufficient evidence of that authority.
Integrity-related grounds concern established misrepresentation, fraud or other legally significant integrity issues under the applicable standard. Suspicion, anomaly detection, investigation referral and adverse factual findings belong to different stages. A fraud risk score cannot be relabeled as proof of fraud. Any explanation must balance legitimate investigative confidentiality with applicable rights to know and contest the substantive ground.
Coordination and administrative grounds concern transaction duplication, allocation of responsibility, billing identifiers and submission formats. They should be marked as correctable workflow events where appropriate. A transaction rejection that leaves the same underlying entitlement open must not automatically become an adverse benefit determination in consumer reporting.
A taxonomy can support multiple grounds. Some decisions depend on a conjunction of conditions, others on independent alternative grounds. A primary operational code is convenient for routing and aggregation, but does not establish that secondary grounds were legally irrelevant. The record should distinguish operative grounds from contextual observations, and identify whether a ground independently supports the result. This avoids an unexamined accumulation of boilerplate reasons that materially changes what a claimant must challenge.
Code design, versioning, and interoperability
A usable code has a stable identifier, a defined meaning, inclusion and exclusion criteria, a domain of applicability and a relation to other codes. Stability concerns identity, not permanent interpretation. When a definition changes, the organization must preserve the version applied at the date of decision. Retired identifiers remain interpretable for historical audit.
A hierarchical taxonomy may begin with broad groups such as eligibility, coverage, procedural requirement and payment adjustment, with subordinate distinctions for particular adjudication issues. Granularity involves a real tradeoff. Excessively broad categories suppress information about why requests fail; excessive detail creates inconsistent manual assignment and sparse observations. A hierarchy allows the published report to aggregate fine-grained reasons while the claims file retains a more specific basis.
Code interoperability requires mapping between vocabularies. A provider-facing billing code and an insurer's internal contract-ground code can describe different facets of one transaction. For example, the CMS remittance coding description distinguishes adjustment reason and supplementing remark information from provider-level adjustments. Mapping these fields onto an adverse benefit determination requires the specific claim context. A crosswalk can be one-to-many, many-to-one or unresolved. Forcing every source code to a single destination creates artificial information where the source has none.
A versioned dictionary entry should record the identifier, label, description, effective dates, related terms, allowed decision types, required evidence fields, applicable jurisdiction or product families, owner and change history. Maintain the exact official external identifier separately where a regulatory or transaction code is used. Obtain authoritative definitions and licenses where necessary; a reconstructed unofficial list should never be presented as the licensed standard.
Free-text explanations and structured codes serve different purposes. Free text preserves contextual detail and can respond directly to an affected person's question; structured fields enable validation, search, statistics and consistent routing. The organization's record can bind them through a decision-event identifier and a set of supporting evidence references, with language versions and templates traceable to the original. This arrangement also makes later corrections observable.
Use a residuary category only where an established determination cannot be assigned a defined substantive category. Require a case-specific explanation and route recurring cases to vocabulary maintenance. Distinguish genuinely novel issues from incomplete investigation, staff convenience, missing data or a software fallback. An increasing share of residual codes is a useful diagnostic, but it does not establish that the underlying decisions were wrong. Its interpretation requires a stable denominator and investigation of the records.
Contractual grounding and evidentiary sufficiency
The reason for a denial requires both an applicable basis and a fact or set of facts that connects the basis to the particular claim. A policy clause alone provides a general rule. A factual allegation alone supplies no authority for a coverage determination. The decisive work lies in establishing the relationship between them.
The applicable basis should be recorded with its source, version, effective date, policy or plan identifier, relevant endorsements or modifications, and the clause or rule actually applied. Claim-specific evidence should be identified with acquisition date, provenance, type, relevance and any material uncertainty. When external regulations modify the contract or affect the consequences of a defect, that legal relationship must be assessed within the appropriate jurisdiction.
In the NAIC model settlement practices, reasonable investigation and a reasonable and accurate explanation appear as related conduct concerns. The institutional significance is that a polished letter cannot compensate for a defective underlying investigation. The FCA (ICOBS 8.1) similarly makes fair handling a distinct concern from settlement speed. These are comparative reference points, not a method for deriving identical legal duties across systems.
Consider a hypothetical claim for water damage under a policy that excludes deterioration but covers a specified sudden accidental discharge, subject to an applicable endorsement. An inspection finds longstanding corrosion and new interior damage. The adjudicator must identify the actual disputed component, distinguish the corroded item from consequential damage, verify the endorsement and evaluate the competing causal descriptions. A label such as 'wear and tear' is insufficient to reproduce the reasoning. The explanation should state which costs are affected, the finding that supports that disposition and the specific provision as applied. The example assumes the stated terms; it says nothing about the outcome of real litigation.
Evidence uncertainty demands explicit treatment. If the decision depends on a document that has not been received, the record should say what information is missing, why it matters and whether the determination remains pending, can be corrected by submission or is final under the applicable procedure. If two reports conflict, record the conflict and the reason one was relied upon. The institutional need for a decision does not eliminate uncertainty about the event.
Evidence confidence, proof requirements and contractual coverage are distinct. A model can estimate the statistical likelihood that a submission is irregular. An expert can report an observed physical condition. An adjudicator may determine whether the established facts meet a contractual requirement. The transformation from probability or observation to adverse determination must remain inspectable, particularly where a claimant can supply contradictory evidence.
Privacy and information access constrain what may be disclosed. Some claim files contain medical information, third-party data, privileged advice or protected investigative details. The system should preserve the underlying reasoning for appropriately authorized reviewers while communicating the permissible substantive explanation to the affected party. Legal exemptions and disclosure restrictions must be established specifically, rather than inferred from the insurer's preference for confidentiality.
Full denial, partial denial, and financial reconciliation
A full denial concerns all components included in the declared entitlement unit. A partial denial concerns at least one component while another is accepted or remains otherwise payable. A partial payment may coexist with pending issues without constituting a final partial denial. Clarifying the unit is essential because a service line may be fully denied within a claim that is mostly paid.
Financial reconciliation is a control against semantic distortion. Let a hypothetical submitted sum be decomposed into paid amount P, recognized contractual cost sharing C, ordinary pricing or payable-limit adjustment L, denied amount D, pending amount U and other explicitly defined adjustment A. A bookkeeping identity can be written as submitted amount = P + C + L + D + U + A, provided the categories are defined mutually exclusively for that report and amounts are comparable. The identity does not establish a legal interpretation of L or A. If an applicable limit is itself disputed as an adverse benefit determination, the local reporting definition may place its affected amount within D instead. Define the classification before aggregation.
For example, take a $1,000 health-service bill with a $700 agreed allowed amount. Suppose $500 is paid, $200 is ordinary patient cost sharing and $300 is a contractual adjustment to the billed charge. No denial is inferred simply from those numbers. A second claim may have an equivalent payment difference because $300 of the submitted sum was found outside the covered entitlement. Identical payments can therefore represent very different adjudications.
A denial notice for a mixed claim should distinguish accepted, adversely determined and unresolved items, state their amounts where known, and explain each materially different ground. Where the final loss amount is uncertain, the notice should identify the determination made and the items for which amounts remain open. This structure assists claimants and enables reconciliation with financial systems.
The amount of a denial should not be inferred automatically from reserves, initial submitted amounts or the difference between a demand and a negotiated settlement. Those quantities are produced for different purposes. Portfolio accounting recognizes liabilities and estimates future payment; claims explainability concerns the contemporaneous adjudication of a specific request. An insurer may have a reserve on a denied claim due to litigation exposure, or a reserve of zero on a pending low-value claim. Neither establishes the correctness of the denial.
In reinsurance and subrogation contexts, multiple contractual relationships may generate different determinations concerning the same underlying loss. A reinsurer's refusal to reimburse a cedant does not by itself establish that the original policyholder's claim can be denied. Each determination requires its own contractual basis, parties and explanation audience.
The structure of a claimant-facing explanation
A useful explanation answers the questions that the affected party can reasonably be expected to have: what was decided, which part of my request is affected, which facts and terms were used, how do those facts relate to the stated ground, what can I do next, and what dates or limits apply? It should identify the particular decision rather than merely describe general claims policy.
The NIST report by Phillips et al. (2021), Four Principles of Explainable Artificial Intelligence, distinguishes the presence of an explanation, its meaningfulness to the recipient, its accuracy regarding the process and the system's knowledge limits. These distinctions are valuable beyond AI. A notice can be meaningful in plain language but inaccurate about why the actual system made its decision. It can accurately reproduce a rule trace but remain unintelligible to its intended reader. It can describe a definitive result without acknowledging that relevant evidence is missing. Evaluating each dimension separately helps locate a defect.
Miller's (2019) social science review of explanation describes explanation as often contrastive, selective and social. In insurance, contrastive questions might include why this component was denied while another was accepted, why an earlier submission was paid, or why the evidence provided did not satisfy the stated criterion. A useful answer selects the material distinction without sacrificing fidelity. It should avoid communicating a speculative counterfactual as a promise that changing one fact would guarantee payment.
A practical notice has a layered structure. The opening identifies the outcome and affected request. The next layer specifies the operative basis in accessible language and identifies the applicable contract or regulatory provisions. A case-specific layer explains how evidence relates to the result, including uncertainty where material. A final layer identifies correction, reconsideration or review options, requirements and applicable deadlines. Supporting documents or technical details may be made available through a suitable secure channel. Requirements imposed by the governing law may demand more than this structure.
For illustration, a notice that merely says 'coverage not established — code X' conveys little. A more useful hypothetical notice states that the request for replacement of an identified item is denied under a specified deterioration exclusion, identifies the inspection evidence, explains why the conclusion applies to that item, distinguishes the separately accepted consequential repairs, and states the specific review route. It should be checked against the actual record before dispatch. If the operative ground was in fact absence of an endorsement rather than deterioration, the latter explanation is wrong even if it sounds plausible.
Language quality includes semantic precision, coherent sequencing, accessible terminology and appropriate translation. A grade-level readability score may detect lengthy sentences, but it cannot establish understanding of the contract or facts. Over-simplification may hide a qualification. Automated translation may alter an exclusion or deadline. Evaluation should test whether the target reader can identify the decision, its basis and the next available action, not merely whether the text is short.
Comparative law and supervisory design
International analysis of denial explanations requires careful separation of legal instruments. The IAIS Insurance Core Principles establish an insurance-supervision framework and a claims-handling benchmark. They do not independently confer identical remedies upon individual claimants. The implementation of conduct standards, the powers of a supervisor, and avenues of private redress depend on local law.
The U.S. NAIC model of 1997 addresses patterns or practices of unfair claims settlement in model statutory terms. State adoption, frequency requirements, penalties, private remedies and procedural obligations can differ. A state-specific claims manual should cite the applicable enacted statute and regulation rather than a model alone. A general-insurance code catalog cannot establish compliance simply by assigning a broad model-law category.
U.S. employee-benefit adjudication illustrates a more detailed domain-specific notification regime. Under 29 C.F.R. § 2560.503-1, adverse benefit determination procedures encompass specific reasons, relevant plan provisions and applicable review information. Paragraphs (g), (h) and (j) differentiate initial notice, review and determination on review. Group health and disability claims have specialized provisions. For example, the group-health review provisions in paragraph (h)(3) include at least 180 days to appeal an adverse benefit determination. That timeframe must not be copied into other insurance lines as a universal deadline. The regulation also contains relevant rules concerning additional information, clinical grounds and urgent or disability cases; exact application requires identification of the plan and claim type.
In the United Kingdom, FCA ICOBS 8.1.1 addresses fair and prompt processing and unreasonable rejection. Associated rules and guidance concern progress and settlement. As a regulatory control, this is related to the quality of operational reasons, even when an exact standardized reason code is not prescribed by that provision. Contract law, applicable regulatory scope and dispute arrangements must be identified separately.
Australia's Corporations Act section 766G, examined in the July 1, 2025 compilation, includes accepting or rejecting claims in whole or in part and assessing insurer liability within the concept of claims handling and settling services. This illustrates a statutory recognition of adjudication as an institutional service rather than only a private transaction. The historical compilation does not establish the currently applicable wording for a particular case in November 2026. A separate legal verification would be required for operational deployment.
The European Union AI Act represents a technology-specific rather than a comprehensive insurance claims instrument. The original Regulation (EU) 2024/1689 Annex III risk classification includes certain life and health insurance risk assessment and pricing applications. That provision should not be read as automatically classifying every claims system as high-risk. Article 86 deals with a conditional right to explanation of certain individual decisions based on high-risk system outputs, within its statutory scope and exceptions. Determining applicability requires assessment of the system's intended purpose, decision and legal conditions, together with the applicable version and application dates. General claims and contractual explanation obligations must still be assessed under their own law.
Across these examples, three informational concerns recur: identifying the grounds, linking those grounds to the actual case, and facilitating correction or review. Their precise legal status and scope differ. A multinational insurer may maintain a common evidentiary data structure and jurisdiction-specific notice rules; it cannot safely substitute a generic global deadline or template for the required local analysis.
Healthcare remittance and administrative data
Health insurance provides an unusually rich setting for distinguishing transactions from determinations because coverage, authorizations, provider contracts, coding conventions, eligibility and benefit payments interact. Provider-facing remittance is often designed for reconciliation and automated processing; a patient-facing benefit explanation serves a different audience. The same record may support both outputs without making them semantically identical.
CMS describes the X12 835 remittance transaction and its use of claim adjustment reason codes and remittance advice remark codes, together with adjustments at the provider level. These components are not equivalent to a complete claimant-facing rationale. A code may explain a payment adjustment while leaving unresolved the contractual or medical justification for a denial. An analyst should identify the transaction set, code set version, adjustment group, claim or service-line unit and supplemental messages before classifying payment events.
A claim rejected by a transaction gateway may never have undergone substantive benefit adjudication. A corrected claim can be resubmitted and paid. A denial of prior authorization occurs before the requested service and may be consequential even when there is no later claim for payment. A denied payment claim concerns a different stage. Combining these events without a distinction can conceal both access barriers and successful technical correction.
CMS's Marketplace public-use files include transparency-in-coverage data. The 2026 file release includes reporting on 2024 experience. Such administrative reporting provides a basis for aggregate research but does not create direct access to clinical files, complete policy texts or case-level review evidence. Reporting units, data availability, coverage of plans and coding instructions delimit the inference.
Long, Lo and Pestaina (2026), in their analysis of 2024 Marketplace denials, report denial rates of 19% for in-network and 37% for out-of-network claims among the HealthCare.gov plans studied. Their analysis also reports that administrative and residual reasons account for substantial shares of reported reasons in the applicable returning-plan subset, and that fewer than 1% of denied claims were appealed. These are aggregate observations from the population and definitions studied, not general rates for all U.S. insurance. The analysis does not establish the correctness of the individual denials.
The distribution of reasons is a distribution of reported classifications, potentially influenced by product structure, administrative practice and reporting rules. A high residual share limits substantive inference. A high administrative share may reflect correctable processing defects, ordinary submission problems or other mechanisms that cannot be distinguished from the aggregate category alone. An apparent change in denial frequency may be caused by revised reporting definitions or code mapping rather than a shift in adjudication.
These data demonstrate why operational denial statistics should retain the reporting unit, time period, network status, plan population, denominator definition and relationship between reasons and distinct claims. The underlying dataset has not been independently reanalyzed for this dossier, and no original frequency estimate is claimed.
Independent evidence of adjudication error
An administrative audit can examine a question that aggregate denial statistics cannot answer: whether the operative decision is supported under its relevant criteria. In 2022, HHS OIG published an audit of Medicare Advantage denials based on stratified random samples of 250 denied prior authorization requests and 250 denied payment requests from 15 large Medicare Advantage organizations, for the period June 1–7, 2019. The report found that an estimated 13% of reviewed authorization denials met Medicare coverage rules, and an estimated 18% of reviewed payment denials met Medicare coverage and organization billing rules. The report describes problems such as overlooked documentation and manual or system-related errors.
These percentages concern denied requests within the audit's defined sampling and review framework. They are not the proportions of all requests denied incorrectly, do not characterize every Medicare Advantage organization, and do not establish the current prevalence of such errors. The distinct samples should not be pooled as though authorization and payment adjudication were identical.
The study is important to explainability because the stated reason may reflect an incorrect representation of evidence. A decision can be internally consistent with the information entered into a system even when a material document in the actual file was overlooked. It can accurately identify an automated edit but fail to justify the resulting refusal under the governing entitlement. Auditing the accuracy of the explanation therefore requires two checks: whether the explanation faithfully reports the actual decision process and whether that process produced a substantively supportable determination.
An independent file review should preserve the original decision version and use only information available at the relevant time when assessing contemporaneous adjudication. A second analysis may ask whether additional evidence subsequently received changes the entitlement. Mixing these questions overstates original error where later evidence was necessary, or understates error where the evidence had already been supplied but was ignored.
The audit evidence also exposes the limitation of appeals as a complete quality signal. Some errors will never be appealed. Some overturned determinations will reflect new evidence, and some upheld decisions can remain disputed. A review program that samples only appealed cases observes a selected population. Independent sampling of accepted, denied and pending cases is necessary to estimate broader adjudication reliability under a defined reference standard.
Human and automated adjudication
Claims systems can use deterministic eligibility rules, statistical models, anomaly detection, document processing, predictive estimates and language generation. These tools differ in their role: extracting evidence, prioritizing review, applying a rule, assisting an adjudicator, rendering a determination or producing a notice. The fact that a system is described as automated claims processing does not identify which of these functions it actually performs.
A deterministic rule engine can record the predicates evaluated, their input values, the applicable rule revision and the resulting branch. But a reproducible execution trace proves only what was computed. A rule may be incorrectly translated from coverage language, an input value may be wrong, or a valid exception may be omitted. Explainability requires both a faithful execution trace and the assessment of substantive applicability.
A predictive model estimates an outcome under assumptions derived from data. A high probability of irregularity or predicted cost is not itself evidence that a contract excludes a claim. When a predictive score prompts investigation, the reason for an eventual adverse determination should state the established operative basis, not substitute the score for evidence. Preserve a separate record of the model's influence on routing or scrutiny where that influence is material to audit and appropriately accessible.
Rudin (2019), interpretable models for high-stakes decisions, argues for considering intrinsically interpretable models where high-stakes choices would otherwise rely on opaque predictions and post-hoc explanations. This is pertinent to claims adjudication, especially when explicit eligibility criteria and readable rule structures can represent the intended decision. It is not proof that a simpler model will always perform better or that interpretability eliminates error. Model suitability depends on task, validation, data and consequences.
Post-hoc feature attribution answers a model-specific question about influential inputs or local prediction behavior. It does not establish the legal relevance of those inputs or their causal effect on the insured event. A feature's attribution can change with the choice of baseline, representation and correlated inputs. An explanation that says a claim was denied because a feature had high attribution can therefore misrepresent the actual contractual decision. Preserve model diagnostics as technical evidence, and translate to an operative denial reason only after establishing the connection to the applicable decision basis.
Generative language systems introduce a further distinction between the reasoning record and the rendered explanation. A language model may produce fluent notices that insert an excluded clause, invent an evidentiary finding or omit an accepted component. A safer organizational design constrains generated content to validated decision fields and source passages, flags unknown values, checks dates and amounts deterministically, and routes discrepancies to authorized review. The model must not silently determine a new ground during letter drafting.
The NIST (2021) explanation accuracy principle separates accurate representation of model process from the correctness of its outputs. This distinction is decisive for insurance: a technically faithful explanation of an invalid underwriting or claims rule still describes an invalid determination. Conversely, a correct determination can be accompanied by a misleading description of how it was reached. Both defects deserve independent correction.
Human accountability requires authority as well as nominal participation. A reviewer should have appropriate training, time, access to evidence and power to correct the system's recommendation. A signature appended without substantive review provides limited assurance. Records should identify the reviewing role, date, decisions overridden or retained, and the reason for material intervention. The appropriate disclosure of individual identities depends on organizational and legal requirements.
Appeals, correction, and procedural fairness
An explanation is practically valuable when it allows the affected party to recognize an error, provide relevant information, obtain reconsideration or use the applicable dispute mechanism. The appropriate pathway varies among first-party insurance, employee benefits, public insurance programs, liability coverage and provider payment contracts. A notice should not imply that one generic route applies to every recipient.
The U.S. Department of Labor claims procedure regulation illustrates how initial notice and full and fair review provisions can structure access to redress in an identified benefit-plan context. The IAIS ICP 19 claims guidance connects explanations and dispute information in a broader supervisory context. Under a relevant regime, the notice may need to identify documents, review instructions, assistance options, timing rules or external avenues. Their exact legal form must be established locally.
Three distinct correction mechanisms deserve separate data fields. Administrative correction repairs a transaction or record error without necessarily revisiting substantive entitlement. Reconsideration revisits the adjudication in light of the existing or additional evidence. Formal appeal or dispute review follows the applicable process, potentially involving a separate decision maker. They can interact, but collapsing them into an undifferentiated appeal code obscures both institutional performance and claimant rights.
For each adverse determination, the organization should preserve the notice dispatch record, applicable route and date, any request for assistance, new evidence, review decision and reasons for reversal or affirmation. Where deadlines are computed by a system, validate triggering dates, calendar logic and relevant legal exceptions rather than relying on a template default. Where communication is inaccessible or incorrectly delivered, assess the effect under the applicable procedure.
A reversal is not automatically proof of original error. It may occur because a previously missing document is supplied, a medical condition changes, the insurer adopts a corrected legal interpretation, a reviewer detects an ignored document or an operational mistake is repaired. Distinct reversal reasons make it possible to discriminate these explanations. An upheld decision is likewise not conclusive evidence that the original adjudication was correct.
A claimant-facing notice should explain what information or issue would be material to a review where this can be identified. An ambiguous demand for more documentation can impose substantial search costs and prolong a legitimate claim. A request for information should be specific enough to permit action, consistent with the applicable legal and evidentiary requirements and respectful of privacy. Avoid requiring a claimant to prove facts that are already established in the organization's own records.
Appeals should be understood as one corrective mechanism within claims governance, not the only filter for accuracy. The KFF (2026) Marketplace appeals findings illustrate that the proportion of denied claims formally appealed can be very small in a defined reporting population. Such a finding makes the interpretation of observed appeals especially sensitive to selection and accessibility. It does not establish why individuals did not appeal.
Measurement design and statistical interpretation
Measurement starts with a precisely declared population. For an operational report, define the insurance line, policy or plan, jurisdiction, time window, decision unit, version of the code dictionary, approach to resubmissions, treatment of pending requests and whether the cohort is assigned by receipt or determination date. Reports with different denominators answer different questions.
Let N be the number of distinct adjudicated claim units with eligible final decisions in a specified cohort; F the number fully denied; and P the number partly denied, with categories disjoint. The full denial incidence is F/N. The incidence of any adverse component is (F+P)/N. An amount-based denial share uses a suitable comparable amount denominator and should be reported separately. Claims with unresolved determinations require explicit treatment.
Suppose a synthetic cohort contains 1,000 final decisions: 120 fully denied, 180 partly denied and 700 with no denied component. Full denial incidence is 12%; any-denial incidence is 30%. Assume that all 300 adverse decisions have 450 operative reason assignments. Dividing 450 by 1,000 gives 0.45 reason assignments per final decision, not a 45% denial rate. A multi-label taxonomy creates exactly this counting problem.
Suppose 90 of those 300 adverse decisions have an administrative reason and 75 have an evidence-related reason, with 25 having both. Their union is 140 decisions, not 165. The individual prevalence values are 30% and 25% of adverse decisions; the union is 46.7%. Reason-assignment shares require a different denominator and should be labeled as such.
For explanation quality, a case-based checklist can include presence of an identifiable disposition, affected component, operative contractual or legal basis, material facts, fidelity to the actual decision record, accessible language, actionable review information and accurate dates and amounts. Some requirements are applicability-dependent, so the denominator for each measure should include only relevant cases. A critical defect involving an incorrect operative ground should not be hidden inside an aggregate score produced by averaging formatting questions.
Measure correctness separately through reference review. If 24 of a simple random sample of 200 denied decisions are found substantively unsupported under a predefined review procedure, the sample proportion is 12%. An approximate Wilson 95% interval is 8.2% to 17.2%, assuming independent observations and simple random sampling. The calculation concerns the conditional population of denials and incorporates sampling uncertainty, not reference-review error. Clustered or stratified data require an appropriate design-based or model-based approach.
The hypothetical audit can be extended to the full population. If 30% of final decisions contain an adverse component and 12% of sampled adverse decisions are unsupported, the implied proportion of all final decisions with an unsupported adverse component is 3.6%, conditional on representative sampling, commensurate units and the same cohort. This product excludes wrong approvals, wrongful delays and other defects. It should not be labeled the total claims error rate.
Appeal statistics require a further denominator. If 30 of 300 adverse decisions are formally appealed, the appeal incidence is 10%. If 12 are reversed, the reversal rate among appealed decisions is 40%; observed reversals account for 4% of all adverse decisions. Neither 40% nor 4% estimates the actual error rate without assumptions about selection and the meaning of a reversal. The data should distinguish new-evidence reversals and original-error reversals.
Missingness is informative. A notice lacking a reason code should be counted as missing rather than silently excluded from the denominator. If a file is unavailable for audit, investigate whether availability varies with severity, claimant characteristics, decision channel or outcome. Report complete-case results alongside sensitivity bounds. If 10 of 100 sampled files cannot be evaluated and 9 of the 90 observed contain a defect, the worst/best-case sample proportions are 19% and 9% if the missing cases are respectively all defective or all sound. A complete-case 10% estimate alone hides this uncertainty.
Comparisons between automated and manual decisions require attention to selection. Automated systems may receive simpler claims while difficult cases are routed to people, or automation may apply to cases with incomplete information. Case mix, product, severity and channel influence both assignment and outcomes. Raw group differences cannot be interpreted as causal effects of automation. A credible evaluation uses a defined intervention, comparable cases, valid assignment or defensible adjustment, and explicit limitations.
Audit methods, reliability, and reproducibility
A defensible audit begins by freezing the applicable code dictionary, sampling frame, decision records and reference standards. The audit question must state whether it concerns decision correctness, explanation fidelity, accessibility, compliance with a defined requirement, or an operational coding defect. These questions can require different reviewers and different evidence.
Stratification may prioritize high-consequence denials, vulnerable claimants, rare reasons, products, decision channels and recent system changes. To estimate a population rate from unequal sampling probabilities, weight observations appropriately. For illustrative strata with 800 ordinary adverse decisions and 200 high-consequence decisions, suppose a simple random sample in each stratum finds defect rates of 5% and 20%. The population-weighted estimate is 0.8 × 0.05 + 0.2 × 0.20 = 8%. An unweighted average of the two stratum rates would be 12.5%, appropriate only under an equal-weight target.
Reference adjudication should be conducted by appropriately qualified reviewers using the operative contract, applicable legal rules, evidence available at the decision time and a documented procedure for contested judgments. Where possible, blind reviewers to the original code or the identity of the deciding channel during the initial assessment. Subsequently compare the reference decision, coded reason and rendered explanation. This separation can reduce anchoring, although complete blinding may be impossible when the file itself reveals prior decisions.
Inter-rater agreement reflects the clarity of the classification task as well as reviewer consistency. Suppose two reviewers classify 100 synthetic explanations as adequate or inadequate: both adequate in 60 cases, both inadequate in 20, reviewer A alone adequate in 10, reviewer B alone adequate in 10. Raw agreement is 80%. Both marginal adequate rates are 70%, so expected agreement under independent marginals is 0.7² + 0.3² = 0.58. Cohen's kappa is (0.80 − 0.58)/(1 − 0.58), approximately 0.524. Agreement is useful information but does not validate the underlying legal or scientific criterion. Resolve disagreements explicitly and assess whether ambiguity in the rubric or the case caused them.
Consumer comprehension should be assessed with tasks, for example asking a participant to identify which item was denied, why it was denied, which evidence could matter and what the next procedural step is. Include members of the intended population, appropriate accessibility accommodations and language variants. Measure comprehension and appropriate action, not only satisfaction. No actual comprehension trial was performed for this dossier.
For automated notices, test correspondence between each generated claim and a verified structured field or source document. Use deliberately constructed edge cases: missing dates, multiple reasons, alternative policy versions, partial coverage, contradictory evidence, corrected submissions and protected data. Test data minimization and the ability to stop generation or request review. A test suite must include adverse cases and failures rather than only ideal complete inputs.
Reproducibility requires retaining dictionary versions, sample selection rules, excluded records, code mappings, reviewer instructions, adjudication dates, measurement scripts and uncertainty assumptions. Confidential records may require secure access and de-identification. Publication of aggregate results does not require publication of personal claim files. The methodology should make clear which components a second authorized investigator could reproduce.
Fairness, incentives, and institutional control
Denial reason distributions can be influenced by incentives at several levels. A claims organization may seek to reduce processing time, administrative costs, invalid payments, litigation risk or losses. Legitimate objectives can produce undesirable behavior if a performance indicator rewards dispositions without measuring correctness or downstream effects. A workflow that rewards rapid closure may encourage premature adverse determinations; a code-volume target may encourage broad residual classifications.
Fairness assessment involves comparison of relevantly comparable claims while preserving the legal relevance of genuine coverage differences. A disparity in adverse determination rates may arise from case mix, benefit design, barriers in documentation, geography, language access, provider practices or discriminatory processes. The observed association alone does not determine its cause. Investigate the grounds themselves, correction opportunities, independently assessed accuracy and the quality of access to review.
Where legally and ethically appropriate, aggregate comparison may use stratification or adjusted models. Avoid conditioning reflexively on variables that are consequences of the process being audited. For example, completeness of documentation may partly reflect accessibility of the insurer's request procedure. Treat it as an explanatory variable only after clarifying the causal question. Small subgroup estimates require adequate uncertainty reporting and privacy safeguards.
Institutional control should separate the authority to establish policy terms, translate those terms into business rules, adjudicate claims, validate models, review adverse decisions and assess market conduct. Separation may be implemented through roles rather than completely distinct departments, but conflicts of interest and escalation authority must be explicit. Significant code changes and automated decision-rule changes should receive review against applicable coverage and law before deployment.
Audit logs should preserve the sequence of evidence intake, decisions, corrections and notices. Access controls protect sensitive material while allowing appropriately authorized independent review. Outsourced operations require contractual and practical oversight: the organization must be able to obtain records, examine the actual decision process, verify notice quality and address failures. This is consistent with the IAIS claims outsourcing guidance, which retains insurer responsibility within its supervisory framework.
A dashboard that reports denial incidence, residual-code prevalence, processing time, complaint counts and appeals separately can help detect anomalies. Its value increases when accompanied by representative file reviews and validated explanation checks. A dashboard cannot determine from aggregate data alone whether an adverse determination is valid. The quality program must maintain a route from any aggregate anomaly to the underlying evidence and decisions.
Error correction generates operational learning. Repeated reversals due to overlooked documentation may indicate ingestion or indexing problems. Repeated code reclassification without substantive reversal may indicate vocabulary problems. Repeated notices naming obsolete clauses may indicate policy-version or template defects. Different corrective actions follow from these findings; undifferentiated denial reduction targets cannot distinguish them.
Implementation across insurance lines
A common information architecture can be implemented without forcing all insurance products into a single legal vocabulary. Property and casualty claims require attention to peril, causation, insured property, valuation and allocation of loss. Life claims may turn on identity, policy status, insured event, beneficiary rights or material representations, as governed by the contract and law. Disability claims can involve multi-period entitlement and definitions of occupational or functional incapacity. Health claims can involve service eligibility, authorizations, benefits, fee schedules, clinical criteria and payment transaction rules.
Liability insurance presents particular distinctions between the duty to defend, indemnification, settlement authority and allocation of covered and uncovered allegations. A third-party claimant may have different information or procedural rights from the insured. A notice must identify the relationship and determination at issue. A coverage denial directed to an insured should not automatically be treated as a resolution of the third party's underlying claim against that insured.
A staged implementation starts with discovery of existing decision fields, code dictionaries, notice templates and legal rule inventories. Identify semantic collisions and missing distinction between denial, rejection and pending states. Establish reviewed mappings from actual local vocabularies to a common analytical layer. Preserve legacy raw fields and versions to avoid falsely rewriting historical decisions.
Next establish a decision record with the unit, outcome, operative grounds and relationship among them, contract version, evidence references, determination date, financial allocation, accountable role and review pathway. Where automation is involved, include the relevant rule or model version, inputs material to the operative process and reviewer interventions. Tailor retention and access to actual law and risks.
Validate notices on representative files before relying on their structure. Reconcile claimed and paid amounts, confirm operative clauses, cross-check dates, test comprehension and inspect differences across languages or accessibility formats. A second-level review is particularly useful for novel grounds, substantial partial denials, conflicting evidence and adverse outcomes that substantially affect access to services.
Introduce monitoring with a baseline and decision-version tracking. Compare changes in code distribution and quality against changes in product, claim population and processing rules. Investigate increases in missing or residual categories before interpreting them as conduct trends. Where adverse changes follow a system update, preserve the prior version and reconstruct the affected cohort.
This implementation is a synthesis of existing claims governance, documentary controls and measurement practice. It is a practical research interpretation, not a certification standard or a new legal doctrine. Each insurer and jurisdiction requires its own coverage, procedural, privacy and supervisory assessment.
Scientific synthesis and research limitations
The comparative investigation reveals that a denial code becomes analytically useful only after the underlying determination unit is defined. Payment messages, claim decisions and coverage grounds possess different semantic content. Standardized codes make large-scale processing and measurement possible, but the compression of case-specific evidence is unavoidable. A reason dictionary should therefore preserve a path back to the underlying adjudication rather than be expected to contain the entire explanation.
Explanation accuracy has two distinct objects. Process fidelity concerns whether the notice identifies the actual grounds and decision process. Substantive adequacy concerns whether those grounds and the supporting facts warrant the result under the applicable authority. These dimensions interact without being reducible to each other. The distinction follows from the comparison of claims conduct instruments, administrative audit evidence and explanation research. It is an analytical conclusion of this study rather than a claim of scientific priority.
The distinction between reason and evidence also provides a way to organize the evaluation of automated systems. A system can assign an operational category well while making poor determinations; it can produce a fluent explanation without identifying the actual basis; and it can be correct in a particular case without having adequate controls for repeatability. The design of evaluation should separate those claims, then assess how they interact in practice.
The institutional significance of an adverse determination lies not only in an individual payment decision but in its consequences for access to correction and the organization's ability to discover systematic error. Aggregate reason counts inform governance only when the counting units, classification versions and data limitations remain visible. Appeals provide valuable evidence about contested decisions, but the selection mechanisms that determine who appeals prevent simple generalization to all denials.
Further research can test reason assignment reliability across lines and jurisdictions, compare consumer understanding of alternative notice structures, estimate the effect of independently validated document retrieval on original-error rates, and evaluate the effects of review procedures on different claimant populations. These are research questions, not completed experiments. Such work requires lawful data access, valid comparison designs and precise outcome definitions.
The source set is selective rather than systematic. It combines international supervisory guidance, selected national legal texts and public administrative literature. The original EU AI Act text and the dated Australian legislative compilation are used within explicit temporal boundaries. No conclusion is offered about their latest consolidated provisions or applicability to individual cases without further verification. Proprietary code dictionaries, individual insurer files and participant-level comprehension data were unavailable. The published U.S. administrative findings were evaluated within their stated study settings without reanalysis of original microdata. The quantitative examples are synthetic and illustrate established statistical concepts rather than observations of industry performance.
No independent external peer review was undertaken. The manuscript's contribution is original organization and comparative interpretation of documented knowledge, with transparent boundaries around the evidentiary basis. It proposes practical methods for evaluating claims explanation and coding quality without introducing a named theory, novel statutory obligation or untested causal claim.
Conclusion
Claims denial reason codes are compact representations of an adverse decision's operative grounds. Their scientific and operational value depends on semantic precision, stable identity, controlled versioning and traceability to a defined determination unit. A code dictionary supports consistency and aggregate research when dispositions, evidence status, financial adjustments, workflow events and review outcomes are represented as distinguishable objects.
Explainability quality concerns whether the affected party and authorized reviewers can identify what happened, why it happened, which evidence and provisions were applied, and how the determination may be corrected or challenged. Its assessment requires fidelity to the actual decision process, substantively adequate grounding, comprehensibility and appropriately actionable information. These dimensions must be evaluated even when the underlying system has an apparently satisfactory predictive accuracy or processing speed.
International supervisory instruments and selected domestic regulations illustrate the relationship between fair claims handling, specific reasons and access to review, while preserving the differences in legal authority and scope. Healthcare remittance illustrates the value and limits of standardized operational messages; public reporting exposes the importance of denominator definitions; administrative audits demonstrate why the adequacy of an explanation must be examined in relation to the underlying evidence.
Institutional accountability depends on preserving a dated, reconstructible record across human and automated operations. Where errors occur, correction must reach both the decision and the communicated explanation. Where a system is changed, subsequent comparisons should account for altered code definitions and case populations. Where claims data are published, the figures should state exactly what has been counted.
The central finding is that the quality of insurance adjudication is strengthened when the ground, evidence, applicable authority, communicated explanation and route to correction remain connected throughout the life of a claim. The resulting information can support both legitimate contractual performance and reliable scientific measurement. This is the practical convergence of claims science, insurance market conduct and consumer protection.
References
International Association of Insurance Supervisors (IAIS). Insurance Core Principles and Common Framework for the Supervision of Internationally Active Insurance Groups. ICP 19, Conduct of Business, standard 19.10 and accompanying guidance. Online consolidated text. Source
National Association of Insurance Commissioners (NAIC). (1997). Unfair Claims Settlement Practices Act. Model 900, January 1997, sections 3–4. Source
Financial Conduct Authority (FCA). Insurance Conduct of Business Sourcebook, ICOBS 8.1, Insurers: general. Rule 8.1.1 and associated guidance. Source
United States Department of Labor. 29 C.F.R. § 2560.503-1, Claims procedure, especially paragraphs (g), (h) and (j). Electronic regulation reproduced by Cornell Legal Information Institute. Source
Centers for Medicare & Medicaid Services (CMS). Health Care Payment and Remittance Advice. Medicare electronic billing guidance. Source
Centers for Medicare & Medicaid Services (CMS). Marketplace Public Use Files. Transparency in Coverage Public Use File, plan year 2026. Source
Long, M., Lo, J., & Pestaina, K. (2026, March 24). Claims Denials and Appeals in ACA Marketplace Plans in 2024. KFF. Source
U.S. Department of Health and Human Services, Office of Inspector General (HHS OIG). (2022, April). Some Medicare Advantage Organization Denials of Prior Authorization Requests Raise Concerns About Beneficiary Access to Medically Necessary Care. Report OEI-09-18-00260. Source
Phillips, P. J., Hahn, C., Fontana, P., Yates, A., Greene, K. K., Broniatowski, D. A., & Przybocki, M. A. (2021). Four Principles of Explainable Artificial Intelligence. NISTIR 8312. National Institute of Standards and Technology. DOI: 10.6028/NIST.IR.8312. Source
Miller, T. (2019). Explanation in artificial intelligence: Insights from the social sciences. Artificial Intelligence, 267, 1–38. DOI: 10.1016/j.artint.2018.07.007. Author manuscript on arXiv, first submitted 2017. Source
Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1, 206–215. DOI: 10.1038/s42256-019-0048-x. Author manuscript on arXiv, first submitted 2018. Source
European Parliament and Council of the European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal L, 12 July 2024. Annex III, point 5(c), and Article 86. Source
Commonwealth of Australia. Corporations Act 2001, compilation dated 1 July 2025, section 766G. Federal Register of Legislation. Source