Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Insured-Life Mortality Experience and Selection Effects

Abstract. Insured-life mortality experience is the observed pattern of deaths and corresponding risk exposure among persons covered under specified life insurance contracts. Its statistical meaning depends on who enters insurance, how they are classified at issue, which policies remain in force, how deaths and exposures are counted, and which demographic population is used for comparison. This monograph examines insured mortality as a conditional and dynamically selected outcome. It distinguishes entry selection, underwriting classification, duration since underwriting, survivor selection, lapsation, policy and benefit weighting, socioeconomic composition, and calendar-period shocks. The research critically synthesizes North American intercompany studies, British and Irish assured-life investigations, international population mortality methods, professional actuarial guidance, and scholarship on informational selection and frailty. Formal derivations establish why insured-to-population ratios cannot by themselves identify underwriting effectiveness or private-information effects; why duration patterns mix several processes; and why policy-count and amount-weighted experience answer different questions. Fully labeled hypothetical calculations illustrate direct standardization, observed-to-expected ratios, competing exits, and uncertainty. The central conclusion is that comparability is an estimand-design problem: each mortality comparison must declare its target population, risk interval, exposure basis, underwriting regime, policy status, calendar years, and adjustment set. The monograph offers a reproducible analytical synthesis and operational research specifications without introducing a proprietary theory or claiming causal identification from inaccessible policy-level data.

Keywords: Insured-life mortality; life insurance; actuarial experience; selection effect; underwriting; select-and-ultimate mortality; duration since issue; adverse selection; advantageous selection; population mortality; standardization; actual-to-expected mortality; competing risks; lapse; frailty; mortality tables; life-years exposed; credibility; insurance risk classification.

Introduction

Life insurance converts the uncertain timing of death into contractual cash flows. A mortality observation becomes actuarially useful only after the insured risk population and the corresponding period at risk have been defined. The life of a person, an in-force policy, a death benefit, and a premium-paying contract are related but distinct observational units. Their counts and denominators lead to different mortality measures. The distinction is fundamental whenever an insurer measures experience, compares itself with an industry table, or translates observations into assumptions for pricing and liability analysis.

The 2015 Valuation Basic Table report documents an extensive U.S. intercompany evidence base: underlying studies covering 2002–2009 incorporated approximately 266 million exposure units by number of policies and nearly 2.6 million death claims across 51 contributing companies. The report developed select-and-ultimate tables and risk-relative subdivisions; it expressly excluded guaranteed-issue, simplified-issue, and pre-need coverage from its specified table-development scope. Its aggregate size gives substantial precision to many descriptive estimates, but participation, definitions, era, and contract mix determine their transferability. A large sample of selected individual lives is not a random sample of the national population.

A different perspective comes from the Human Mortality Database methods protocol, revised August 5, 2025. Population mortality research converts vital-registration deaths and census or population estimates into age-, period-, and cohort-specific death rates and life tables. Life insurance experience starts with selection into a contract and with insurer administration, whereas population experience starts with a defined resident or registered population. Matching the two requires attention to the underlying counting processes, not simply similar-looking age labels.

The intellectual and empirical challenge is visible in the Cawley and Philipson (1999) life-insurance study. Its findings challenge a simple prediction that higher self-perceived mortality risk must accompany greater life insurance holdings. Einav and Finkelstein (2011) likewise show, across insurance markets, why the empirical direction of selection can vary with preferences and risk. These works counsel against treating the existence of underwriting, the statistical composition of insureds, and adverse selection as equivalent findings.

This dossier examines mortality under individual life insurance, with selected comparisons to guaranteed-issue products, group or collective arrangements, and annuitants when those contrasts illuminate selection. The principal object is the mortality of insured persons and contracts, not the entire economics of life insurance, the separate topic of stochastic mortality projection, or annuity valuation. Product-specific eligibility, benefit triggers, exclusions, and statutory rules must be checked within the relevant jurisdiction. The analysis has an international comparative scope and draws concrete evidence from jurisdictions in which accessible studies permit defensible inferences.

The evidentiary method is critical source synthesis combined with transparent mathematical reconstruction. The original analytical contribution lies in specifying comparable mortality estimands, explicitly decomposing observed differences into identifiable and unidentified parts, connecting underwriting duration to survivor and lapse selection, and proposing falsifiable diagnostic tests for experience comparisons. All numerical examples below are constructed to demonstrate arithmetic and identification problems. They are not estimates derived from actual insurer policy records. No new proprietary actuarial theory is asserted.

Definitions, Historical Foundations, and the Research Object

Mortality as a conditional frequency and intensity

Let TT denote the age at death, xx attained age, tt calendar time, dd duration since policy issue, and I(t)=1I(t)=1 an indicator that a specified life insurance contract is in force at time tt. Define the insured force of mortality for a precisely identified risk set as

μI(x,t,d,z)=lim⁡Δ↓0Pr⁡(t≤T<t+Δ∣T≥t,I(t)=1,x,d,z)Δ,\mu_I(x,t,d,z)=\lim_{\Delta\downarrow0} \frac{\Pr(t\leq T<t+\Delta\mid T\geq t,I(t)=1,x,d,z)}{\Delta},

where zz collects measured risk characteristics. Conditional on a constant mortality force within a unit interval, the associated one-year death probability is q=1−e−μq=1-e^{-\mu}. In a real experience study the instantaneous force, central death rate, and annual death probability need not be numerically identical; approximating one by another is justified only within an explicitly described exposure and rate model.

The insurance risk set is contract-defined. For a single-life policy one person can have several contracts, and a death may trigger multiple benefits. A study counting death claims per policy exposure will generally treat multiple claims differently from a study counting distinct deceased persons. A life-years denominator measures exposure of unique persons; a policy-years denominator counts covered contracts. An amount-years denominator multiplies each interval of coverage by the relevant insured amount. Analysts must specify how joint-life policies, multiple policies on the same person, riders, accelerated death benefits, reinstatements, and administrative reporting dates enter the numerator and denominator.

The 2017 Individual Life Insurance Mortality Experience Report illustrates the practical need to compare actual mortality with defined industry tables and to segment results by policy characteristics. Its public description identifies underlying tabulations and large data files; the existence of such files does not establish that the present monograph has independently recomputed them. The 2009–2016 U.S. study specifically concerns standard individually underwritten issues and experience by amount of insurance. These design descriptions are essential restrictions on external inference.

What selection means

Selection has several scientifically distinct meanings. Entry selection is the difference between those who seek or obtain life insurance and the reference population. Underwriting selection is the assessment, acceptance, postponement, decline, or classification of applicants according to available risk information. Risk-class selection refers to the resulting differences between preferred, standard, and substandard accepted lives. Survivor selection occurs because individuals with higher death hazards tend to leave a surviving cohort sooner. Persistency selection occurs when policy lapses, conversions, surrenders, or other exits alter the mortality composition of coverage that remains in force. Information selection concerns differences between what applicants know and what the insurer observes or prices.

The U.S. risk classification practice standard addresses classification characteristics, actuarial relevance, objectivity, practicality, and applicable law. Its professional and jurisdictional scope is U.S. actuarial practice; it does not establish the legality of a particular characteristic in all markets. The purpose of classifying expected insurance costs is conceptually separate from a claim that every observed class difference has a causal origin in the underwriting intervention.

Adverse selection, in an insurance-market sense, arises when private information or other unpriced information influences take-up or coverage choice in a way that worsens the expected insured loss relative to the premium basis. It is a behavioral and informational proposition with a counterfactual pricing component. A selected insured cohort can have lower average mortality than the general population and still contain adverse selection within particular coverage, duration, or benefit cells. Advantageous selection is also possible when insurance demand is positively associated with health-preserving behaviors or risk aversion. The selection review by Einav and Finkelstein provides the theoretical and empirical basis for treating the sign of selection as an observed market question.

Select and ultimate mortality

A select table indexes mortality by age at issue and elapsed duration since issue, reflecting the conditional nature of the accepted cohort. An ultimate table primarily indexes attained age after an adopted select period, often with other classifications retained. Such a table is an actuarial representation of experience and assumptions; its ultimate period does not establish that all original selection effects have literally vanished.

The Academy/SOA 2015 VBT methodology contains both select-and-ultimate and ultimate-only formats, with smoker distinctions and preferred-risk relative tables. Its documentation highlights that historical smoker information was richer in select observations than in older ultimate exposures, requiring an estimated prevalence structure. This illustrates how apparent duration effects can partly reflect historical changes in classification variables and data capture rather than biological changes after issue.

For historical perspective, the International Actuarial Association's 2009 research on insurance liability measurement distinguished insured mortality from population mortality and discussed diminishing underwriting effects over elapsed policy durations. The appropriate substantive interpretation is conditional: strength and persistence of a select effect depend on product rules, underwriting practices, age, risk class, and lapse behavior; a universal select-period length is unsupported.

Boundaries of documentary evidence

The principal contemporary institutional evidence includes the Society of Actuaries Research Institute and LIMRA for U.S. experience; the Continuous Mortality Investigation (CMI) for assured lives in Great Britain and Ireland; the Human Mortality Database and United Nations for population death rates; and actuarial standards bodies for professional procedures. Some research packages contain paid databases or restricted working papers. In those instances the publicly accessible institutional summary supports only the summary-level claim made here, rather than unaudited cell-level estimates.

The distinction between evidence and inference governs the argument throughout. An intercompany table can document a conditional mortality pattern. It cannot, without stronger design, identify which portion arises from underwriting effectiveness, economic selection into purchase, health behavior, wealth, differential survival, coding, or changes to the roster of participating companies. Establishing this distinction is the condition for sound comparative mortality science.

Insured Cohort Selection

Entry selection as a sequence of decisions

The insured population is produced through a sequence: knowledge of insurance, willingness and ability to apply, product and amount chosen, information submitted, insurer eligibility rules, medical or nonmedical assessment, acceptance and issued class, premium affordability, and continued coverage. These steps have different conditioning consequences. Conditioning on issued policies removes rejected applications and unsuccessful offers, while conditioning on in-force policies additionally removes policies that have ended. The resulting insured experience is therefore a study of a specific longitudinal selection process.

Let ZZ represent the vector of determinants of mortality, AA the event of application, UU the underwriting outcome, and FtF_t continued coverage at time tt. A conceptual selection distribution is

f(z∣A=1,U=accepted,Ft=1,T≥t).f(z\mid A=1,U=\mathrm{accepted},F_t=1,T\ge t).

It is distinct from the distribution f(z∣T≥t,population)f(z\mid T\ge t,\mathrm{population}). The accepted insured mortality at a given age and time integrates individual hazards over the selected distribution,

μˉI(x,t)=∫μ(x,t,z) fI(z∣x,t) dz.\bar\mu_I(x,t)=\int \mu(x,t,z)\,f_I(z\mid x,t)\,dz.

Population mortality integrates hazards over a different distribution. A lower crude insured death rate can therefore arise even if the underlying mortality function conditional on every relevant attribute is the same in both groups. This is a standard consequence of mixture distributions, not evidence that the insurer reduced anyone's intrinsic death hazard.

The mechanism is visible in a simple constructed example. Consider two equally defined age-risk strata with central death rates of 2 and 8 deaths per 1,000 person-years. If insured exposure is 80% in the lower-rate stratum and 20% in the higher-rate stratum, the expected insured aggregate rate is 0.8(2)+0.2(8)=3.20.8(2)+0.2(8)=3.2 per 1,000. If the reference population is 60% and 40%, its rate is 0.6(2)+0.4(8)=4.40.6(2)+0.4(8)=4.4 per 1,000. The insured-to-population crude ratio is 3.2/4.4=0.72733.2/4.4=0.7273. Yet the two groups have exactly identical mortality within each stratum. A crude advantage of approximately 27.3% is entirely compositional. Standardizing the insured stratum-specific rates to the population weights gives 4.4, removing this constructed aggregate difference.

Suppose, in a different hypothetical example, the insured stratum rates are 1.8 and 7.2 while the population rates remain 2 and 8. Standardization to the population weights yields 0.6(1.8)+0.4(7.2)=3.960.6(1.8)+0.4(7.2)=3.96 deaths per 1,000, or 90% of the reference rate. The remaining standardized difference is a conditional association within the two coarse strata. It still cannot be ascribed uniquely to underwriting because health, income, occupation, geography, and purchase behavior may remain unmeasured within the strata. That is the essential distinction between demographic standardization and causal identification.

Risk classification and the information boundary

Underwriting may employ health disclosures, medical examinations, prescription or clinical histories, occupation, hazardous activity, and legally permitted external information. The permissible information set, consent requirements, nondiscrimination rules, and privacy requirements depend on jurisdiction and time. Underwriting classifications may be broad, while individual risk is continuous and heterogeneous. Even within a supposedly homogeneous preferred class, residual differences can change as the portfolio ages.

The actuarial standard ASOP No. 12 requires consideration of the relationship between risk characteristics and expected outcomes, practical measurement, applicable law, and testing of classification systems in the American professional framework. A scientific study must distinguish three separate matters: whether a characteristic predicts mortality in observed data; whether it is available and lawful for an insurer to use; and whether using it improves a decision under a specified loss, fairness, and stability criterion. Predictive association does not by itself answer the latter two questions.

The 2017 Guaranteed Issue Mortality Tables provide a useful counterexample to equating all life insurance with traditionally medically selected mortality. Their existence as a distinct table family reflects a different product and entry mechanism. The 2015 VBT report excludes guaranteed-issue experience from its own development scope. Consequently, applying a preferred-underwritten experience curve to guaranteed-issue or final-expense portfolios without credible adjustment would cross an evidentiary boundary.

Coverage choice, hidden information, and preference-based selection

A demand-side selection hypothesis has to specify the variable on which selection operates. Buyers may select whether to purchase, how much face amount to purchase, which product to buy, whether to add riders, or whether to maintain coverage. For each margin, a test of adverse information must distinguish the risk the applicant knows from the information already reflected in the offered price. A positive correlation between coverage and mortality need not identify hidden information because premium schedules, income, wealth, dependents, agency distribution, and insurer screening simultaneously affect both.

The Cawley–Philipson empirical study is especially relevant because it examines life insurance using information on self-assessed and actual mortality risk and coverage. It reports patterns inconsistent with some simple asymmetric-information predictions. The result should not be generalized into universal proof of advantageous selection: its sample, institutional environment, tested margin, and historical period determine the claim. The broader Einav–Finkelstein synthesis clarifies that risk preferences and expected costs can covary in different directions. A risk-averse, prevention-oriented applicant may demand more insurance while also displaying lower expected mortality.

At the contract level, the most defensible empirical question is conditional: after accounting for observable risk class, price schedule, demographic composition, distribution channel, and policy options, does additional chosen coverage predict subsequent mortality, and is that relationship stable across underwriting regimes? Answering this question needs individual-level data, model diagnostics, and a defensible identification strategy. Industry aggregate A/E ratios are inadequate on their own.

Socioeconomic selection and geography

Purchase, acceptance, and persistency depend partly on resources and institutional access. Thus wealth, earnings, education, occupational hazards, neighborhood deprivation, migration, and access to health care can alter the composition of a commercially insured cohort. Cross-country comparisons magnify these issues when insurance penetration, mandatory or voluntary group coverage, and public welfare institutions differ. A national population table averages over persons who may have very different opportunities to obtain individually underwritten cover.

The 2026 CMI term-assurance investigation specifically includes results analyzed by deprivation information; the publicly recorded October 2026 correction to some deprivation and cause-of-claim exhibits demonstrates the need for source-version discipline. Sociodemographic stratification is useful only if the area measure, geography, time period, and coverage selection are represented correctly.

A useful cross-market decomposition therefore begins with age and sex, adds calendar period, smoking or underwriting class when supported, and then considers product, sum assured, distribution, geography, and socioeconomic composition. Standardization should be limited to covariate cells with support in both samples. If an insurer insures a narrow affluent segment that has no comparable national distribution in certain cells, an extrapolated "nationally standardized" rate rests on a model rather than an observed overlap.

Selection, equity, and research interpretation

Scientific comparison requires clear separation of individual predictions, group rates, and access consequences. Differential mortality estimates can be accurate and still expose a trade-off between segmentation, affordability, inclusion, and information governance. The existence of a mortality differential neither requires its use in price classification nor resolves the ethical question of permissible discrimination. Population comparisons can inform distributional analysis, but a policy judgment must specify legal authority, social objectives, and who bears the financial consequences. The dossier uses risk classes to explain measured insured experience, not to recommend particular personal underwriting decisions.

Underwriting, Policy Duration, and Dynamic Selection

Time since underwriting is an explicit time axis

A newly issued policy has an age at underwriting and an attained age at each observation. For a policy issued at age x0x_0, attained age is approximately x0+dx_0+d, subject to exact conventions for birthdays and policy anniversaries. Calendar year and cohort introduce additional axes. A duration gradient can be confused with age or period effects if observations are collapsed. Comparing duration one and duration fifteen without standardizing attained age, issue cohort, and underwriting class may reproduce a change in composition rather than decay of underwriting selection.

A select mortality table represents q[x0]+dq_{[x_0]+d} during selected durations and qx0+dq_{x_0+d} on the ultimate basis under its table conventions. A typical select advantage reflects higher-risk applications not entering preferred cohorts, but its evolution also reflects subsequent deaths, lapses, policy conversions, reunderwriting, new issuances, changes in diagnostic technology, and the passage of time. The 2015 VBT development report documents the structural difference between select and ultimate rates while also revealing important historical changes in available classification data.

The most useful duration study is therefore a set of issue-cohort trajectories, observed under explicit rules for changes in policy status. An analyst should report conditional hazard or A/E at fixed attained age where possible, with issue year and calendar period either controlled or displayed separately. A duration effect is an empirical conditional pattern; a statement about how long underwriting "works" is a causal interpretation requiring additional evidence.

Survivor frailty and the changing risk set

Unobserved risk differences are especially important as a cohort survives. Let an individual conditional hazard equal Zμ0(t)Z\mu_0(t), where the frailty variable ZZ is nonnegative. Individuals with larger frailty are removed more rapidly by death, so the distribution of ZZ among survivors generally shifts toward smaller values over time. Aggregate mortality can decelerate relative to individual hazard paths even without any change in underwriting. This mathematical insight is developed in Vaupel, Manton, and Stallard (1979), a foundational study of heterogeneity and selection in life-table dynamics.

For a simple two-type illustration, suppose half a cohort begins with annual forces 0.005 and 0.015. After ten years of survival with constant individual forces, the higher-risk fraction among survivors becomes

wH(10)=0.5e−0.150.5e−0.05+0.5e−0.15=e−0.15e−0.05+e−0.15≈0.475.w_H(10)=\frac{0.5e^{-0.15}}{0.5e^{-0.05}+0.5e^{-0.15}} =\frac{e^{-0.15}}{e^{-0.05}+e^{-0.15}}\approx0.475.

The initial mixed hazard is 0.010; the survivor-mixed hazard after ten years is 0.525(0.005)+0.475(0.015)=0.009750.525(0.005)+0.475(0.015)=0.00975, although neither individual's hazard has improved. The calculation is deliberately stylized, with no aging trend, to isolate the selection mechanism. In realistic data aging raises baseline hazard; frailty selection changes the observed slope relative to that baseline.

Lapse as a competing exit

A policy disappears from an in-force mortality study after lapse or surrender. Let λD(t∣z)\lambda_D(t\mid z) be the death hazard and λL(t∣z)\lambda_L(t\mid z) the lapse hazard for a policy with covariates zz. The probability that a contract remains active and the insured alive is governed by both decrements:

Sin force(t∣z)=exp⁡[−∫0t{λD(u∣z)+λL(u∣z)} du].S_{\mathrm{in\ force}}(t\mid z)= \exp\left[-\int_0^t\{\lambda_D(u\mid z)+\lambda_L(u\mid z)\}\,du\right].

This expression assumes a continuous-time competing-decrement representation conditional on the chosen covariates. It describes contract survival, not the person's biological survival after lapse. The death hazard observed among those still insured integrates over a distribution continuously altered by both exits.

The SOA post-level term study investigates shocks to lapses and mortality around the end of level-premium periods. Contract economics create predictable opportunities for selection: when premiums jump, policyholders may choose continuation based on health information, affordability, and the availability of replacement coverage. A high lapse rate may leave a materially different cohort in force. Observed post-level mortality can therefore reflect both longer duration since original underwriting and policyholder decisions made after a price change.

The 2024 SOA–LIMRA lapse investigation reports about 135.9 million policies exposed and 5.4 million surrender or lapse events over its study design. Its public summary documents behavior at substantial scale; it cannot alone show whether any particular lapse caused a subsequent mortality difference. A complete analysis would link lapse paths and, where legal and feasible, mortality following termination, with selection and censoring assumptions clearly declared.

Reunderwriting, conversion, and product transitions

A conversion option can allow a term policyholder to move to another product without fresh medical evidence, depending on contract terms and timing. Such an option can attract the subset for whom new underwriting would be costly or unsuccessful. The converted block may therefore be differently selected from both newly underwritten permanent insurance and the original term block. Similarly, reinstatement rules, guaranteed insurability riders, and benefit increases can reset or modify the relevant risk selection date.

Duration since the original issue cannot be the sole time axis after a material contract transition. The analyst should capture issue date, last medical-underwriting date, conversion date, rating class before and after transition, and any change in coverage amount. Otherwise, a cohort called "duration ten" may combine a decade-old underwriting decision with recent ununderwritten increases of coverage.

Why lapse-mortality associations need care

A simple comparison of deaths among retained and lapsed policyholders is usually unavailable because insurance administration ceases observing deaths of many former policyholders. Even where subsequent deaths are linked, retained and lapsed groups differ in health, price elasticity, financial resources, and replacement options. A life insurer can estimate in-force claims experience credibly and still lack identification of mortality had all policies remained active. The relevant counterfactual is not observable merely by recoding lapsed exposures as zero claims.

Practical diagnostics include fixed cohorts before and after premium increases, underwriting-class-specific lapse rates, duration-age interactions, status transitions, and sensitivity analysis for informative termination. Any claim of post-lapse mortality selection should disclose whether external mortality linkage, natural experiments, suitable instruments, or credible longitudinal adjustment supports it.

Population Comparability and the Measurement of Experience

Defining the population contrast

The statement "insured lives die less often than the general population" is incomplete until age, sex, jurisdiction, calendar year, exposure basis, and policy scope have been fixed. An insured death rate can refer to newly underwritten individual term policies; all active individual policies; guaranteed-issue contracts; group life certificates; converted term policies; or persons with any coverage. Each population contains different selection mechanisms. A general-population table may refer to national residents, citizens, registered deaths, or a census-defined denominator. Its death rates include persons never eligible for a particular insurance product.

The 2025 HMD methods protocol records explicit procedures for deaths, population exposures, Lexis time-age representations, and the construction of period and cohort rates. Its August 2025 revision improves population exposures using monthly birth information and updates a method for infant mortality. These are evidence of the precision required in demographic measurement. Insurance experience likewise needs exact birth dates, entry and exit dates, and duration conventions; a difference in denominator construction can produce a difference in estimated mortality even before the groups' health compositions are considered.

The UN World Population Prospects 2024 covers 237 countries or areas and combines available vital statistics with demographic estimation and projection procedures. Countries differ in death registration, census frequency, and estimation uncertainty. Population rates drawn from model-based estimates should be treated differently from rates directly supported by complete registration. The analyst must identify whether an external table is period or cohort, observed or estimated, sex-specific or combined, and whether a pandemic or crisis interval changes interpretation.

Exposure reconstruction and internal consistency

For each covered risk interval [ai,bi)[a_i,b_i), the policy or life exposure contributed to the observation window is its intersection with that window, subject to coverage rules and censoring. A central estimate can use

E=∑i∫window1{i alive, in force, eligible} dt,m^=D/E,E=\sum_i\int_{\mathrm{window}}\mathbf{1}\{i\ \mathrm{alive,\ in\ force,\ eligible}\}\,dt,\qquad \widehat m=D/E,

where EE is measured in person-years or policy-years and DD is the corresponding count of death events under the chosen unit. If the policy count duplicates persons, it must be labeled accordingly. For an interval of fractional duration, the numerator and denominator must treat a death consistently. Central rates approximate hazards for suitably fine cells; conversion to an annual qxq_x requires a within-year deaths or hazard assumption.

The administrative death date, notification date, claim admission date, and claim settlement date can differ. A mortality study on date of death and a claims financial report on date of settlement need not report identical calendar totals. Policies written shortly before the cutoff may also have late-reported deaths. An experience analysis therefore needs a closure convention and either sufficient runoff or an adjustment and sensitivity analysis for incomplete claim emergence.

The Data Quality standard, ASOP No. 23, adopted for U.S. actuarial practice, addresses the selection, review, use, limitations, and disclosure of data. Its distinction between a data review and a formal audit matters: a consistent data file can still omit deaths or misstate exposure. The NAIC's VM-51 resources demonstrate contemporary administrative attention to structured U.S. mortality data collection, dictionaries, and validations, including 2026 materials for the 2025 observation year. These reporting tools support data engineering; they are not proof that an individual insurer's mortality extraction is complete.

Actual-to-expected mortality and the choice of benchmark

A common experience measure is the actual-to-expected ratio,

A/E=D∑i∫μiref(t) dt,\mathrm{A/E}=\frac{D}{\sum_i \int \mu_i^{\mathrm{ref}}(t)\,dt},

where a chosen reference hazard generates expected deaths for the same covered exposures. In annual discrete tabulations, expected claims may be computed from properly exposed policy-years and the matching qq or mm convention. A/E of 85% indicates actual deaths of 85% of expected under that particular table, timing, and exposure rule. It does not mean insured mortality is 15% below population mortality unless the expected basis is a harmonized population table. Nor does an A/E of 100% prove that the experience study or reference basis is unbiased.

Consider a hypothetical portfolio with 10,000 life-years and 100 observed deaths against 120 expected deaths from a specified table. Its A/E is 100/120=0.8333100/120=0.8333. Under a deliberately simple independent Poisson-count approximation with a fixed expected denominator, the approximate standard error of the log ratio is 1/100=0.101/\sqrt{100}=0.10. A log-normal approximation then gives a 95% interval of roughly 0.6850.685 to 1.0141.014. These numbers are a mathematical demonstration, not a fit to insurer data; overdispersion, clustered policies, common shocks, and uncertainty in the reference rates can widen or otherwise alter uncertainty.

A/E changes when the reference table, its mortality improvement projection, or risk-class mapping changes. Comparing A/E values from two studies only makes scientific sense when the table and calculation conventions are aligned or their differences explicitly decomposed. The U.S. 2017 experience study description identifies comparisons to standard industry tables by policy characteristics; the term "expected" consequently refers to an actuarial benchmark, not an immutable counterfactual.

Policy weighting versus amount weighting

A life insurer can evaluate mortality by policy count, unique person count, death benefit, annual premium, or some liability-relevant financial measure. These measures give different importance to lives. For policy-count A/E, the numerator is the count of death claims and the expected number derives from exposed policies. For amount A/E, the numerator is the sum of insured amounts payable on deaths and the expected amount is the exposure-weighted sum of amounts multiplied by applicable death probabilities.

Consider 1,000 hypothetical policies each with a $10,000 face amount and a mortality probability of 0.005, plus 100 policies each with a $1 million face amount and probability 0.002. Expected policy deaths are 1,000(0.005)+100(0.002)=5.21{,}000(0.005)+100(0.002)=5.2; the overall policy-count mortality frequency is 5.2/1,100≈0.004735.2/1{,}100\approx0.00473. Total face amount is $110 million, while expected death benefits are $50,000 plus $200,000, or $250,000. The face-weighted expected claim frequency is 250,000/110,000,000≈0.00227250{,}000/110{,}000{,}000\approx0.00227. Both are internally correct for their respective weights. Neither is a more authentic biological death probability for the population without a specified target.

Differences between lives-weighted and amounts-weighted experience have substantive interpretation: sums assured correlate with socioeconomic composition, underwriting class, age, and product design. They do not establish the direction of underwriting quality in isolation. The CMI Working Paper 215 summary reports a difference between amounts-weighted and lives-weighted term-assurance experience, giving a contemporary institutional example of why the choice of weight matters.

Direct standardization and transport

Let ss index strata and ws∗w_s^* be weights from an explicitly selected target population. The directly standardized insured rate is

mI∗=∑sws∗mI,s,withmP∗=∑sws∗mP,s.m_I^{*}=\sum_{s}w_s^*m_{I,s}, \quad\text{with}\quad m_P^{*}=\sum_{s}w_s^*m_{P,s}.

A standardized rate ratio mI∗/mP∗m_I^*/m_P^* compares rates at the same chosen mixture of strata. The result depends on the chosen target weights, and sparse or empty cells require pooling, modeling, or transparent restriction of the target population. Age adjustment alone does not ensure comparability where insured classes also differ in smoking status, income, occupation, disease history, or entry mechanisms. Standardization can reduce compositional distortion, but it neither repairs unknown covariates nor transforms observational findings into randomized causal effects.

For international comparisons, common strata may need to be coarser to preserve overlap, thereby leaving more residual heterogeneity. Sex/gender variable coding, smoking classification, occupation, migration, issue-age rounding, and medical evidence thresholds can all differ between countries. Cross-country transport must identify a common mortality estimand before fitting a common regression.

Credibility, uncertainty, and model validation

The professional Credibility Procedures standard instructs U.S. actuaries to consider the relevance, homogeneity, and predictive usefulness of subject and supplementary experience. A sparse class-duration cell may contain an extreme observed A/E due to chance. Credibility methods can combine local data with more relevant benchmarks, but the result inherits assumptions about relevance, exchangeability, and stability. Mixing a guaranteed-issue cell with preferred-underwritten experience solely to raise sample size can reduce sampling variance while introducing larger systematic bias.

Uncertainty has several components: random variation in deaths; incomplete exposure or death recording; uncertainty in risk classification; historical change in underwriting rules; table parameter uncertainty; mortality trend variation; and selection or censoring bias. Bootstrap or Poisson intervals address only specified stochastic elements. A model that reports narrow confidence intervals while ignoring changing policy composition presents sampling precision as if it were total scientific certainty.

Modeling practice should include calibration by age, sex and class; holdout validation by issue year and calendar period; residual checks by duration and product; influence analysis for large amounts; an explicit treatment of data leakage; and documentation of model changes. The U.S. actuarial standards body identifies ASOP No. 56 on modeling as covering model design, development, use, evaluation and validation, with effectiveness from October 2020. A model's suitability for one insurer and valuation purpose does not confer global predictive validity.

Cause of death and measurement drift

Cause-of-death coding can help investigate mortality shocks and socioeconomic differentials, but inconsistent diagnostic or coding practices complicate interpretation. The WHO ICD-11 overview describes a standardized framework for recording and comparing mortality and morbidity information. For a life insurer, cause-of-claim fields may arise from claims documentation rather than the same system used for national death certification. Differences in unknown-cause share, medically reviewed claims, and terminal illness benefit treatment can distort cause-specific comparisons.

Underlying-cause analysis also differs from all-cause mortality analysis. An all-cause death claim count may be sufficiently reliable to measure aggregate experience while cause-specific attribution remains incomplete. Comparisons must record the classification version, coding date, missingness, and whether retrospective revisions occurred. Findings about causes should not be used to infer differences in overall mortality without establishing comparable denominators.

Empirical Evidence, Comparative Research, and Stress Tests

United States: industry studies, scope, and interpretive limits

The U.S. 2015 VBT was built to support actuarial valuation tables for individually underwritten life insurance with standard and preferred underwriting, rather than to estimate an unconditional national mortality advantage. The 2015 VBT report's underlying-data section records four intercompany study periods between 2002 and 2009, 51 contributors across the compiled data, and a complex reconstruction of smoker prevalence at older policy durations. Two design implications follow. First, a reference table inherits the underwriting and sales environment of its data. Second, a table can be reliable for its approved intended use while being a poor reference for other contract types or contemporary applicant populations.

The 2009–2016 individual mortality report and the 2017 experience report extend institutional experience study beyond the historical inputs of the 2015 VBT. Their published resource descriptions distinguish covered years and data products. An analyst cannot infer a 2026 U.S. underwriting improvement from their existence alone. A robust longitudinal study must examine consistent cells or formally adjust the changing participating companies, age structures, distribution channels, and underwriting practices.

As of the research cutoff, the Society of Actuaries and LIMRA have publicly described a 2018–2024 individual-life data collection that seeks detailed policy information including COVID-period experience and causes of death. The publicly reviewed page is a data-request description, not evidence that a specific resulting mortality estimate has been finalized, released, or replicated. The corresponding program's published-versus-upcoming listings must be distinguished by actual release status. This distinction prevents future-study expectations from becoming false empirical claims.

The guaranteed-issue table series supplies a substantive warning: even within one country, different issue restrictions warrant different mortality evidence. Its publisher documentation also records later corrections to certain spreadsheet values and a prior caching problem. Thus source version, publication date, and amendment history are part of mortality data provenance, not merely bibliographic housekeeping.

United Kingdom: standard and rated assurances

The Continuous Mortality Investigation has established a long-standing institutional program for assured-life experience. Its assurances investigation overview distinguishes term, whole-life, and endowment experience, rated lives, and the "16" Series mortality tables based on 2015–2018 experience for non-rated term policies. "Rated" and "standard" are underwriting categories in that study, not global biological group labels.

The most recent publicly described CMI Working Paper 215 covers term-assurance experience for 2022–2024. Its publisher abstract reports slightly lighter lives-weighted mortality for standard lives than the T16 basis and considerable differences for rated lives; it also describes smoker and nonsmoker subdivisions and a distinction between amount and life weights. The October 7, 2026 revision corrects specified charts and a table concerning deprivation and causes of claim. The publicly available summary does not permit this dossier to reconstruct confidential cell-level experience or independently verify every chart in the restricted working paper. All substantive quantitative comparisons here therefore remain limited to the publicly issued summary.

This evidence offers an important interpretive lesson. A rated classification may appear to have lighter actual-to-expected experience against a table calibrated for a particular basis without demonstrating that a rated individual has lower mortality than a standard individual. Denominator choice matters. A class-relative benchmark, mortality loading, and national population rate are different objects. A scientific statement about the "mortality of rated lives" must identify its comparison base.

Ireland: institutional and temporal differences

The CMI–Society of Actuaries in Ireland study describes term-assurance experience for 2015–2021. Its public high-level results report mortality close to the T16 benchmark with modest variations by calendar period, including a different direction in 2020 from 2021. The source is an accessible institutional abstract for a restricted full report; the figures should not be treated as independently reproduced population mortality.

A direct Ireland–Great Britain contrast would require common age and sex definitions, smoker and risk-class groupings, comparable T16 mappings, and exposure units. The studies differ in period and contract mix, so observed A/E differences are insufficient for a jurisdictional ranking. A more defensible comparison would reweight to a shared demographic distribution and quantify remaining uncertainty separately for each market.

Population mortality as a comparison, not a replacement

The Human Mortality Database protocol and UN population estimates provide ways to define national mortality references. Such references include individuals with and without private life insurance, and include social and geographic populations with different coverage availability. Population life tables are indispensable for background rates and large-scale mortality shifts but are not ready-made "expected" tables for every insured portfolio.

A country with nearly universal group protection, a country with predominantly voluntary medically underwritten cover, and a country where private life insurance is restricted to a small income segment can have similar biological age-specific rates yet very different insured-to-population ratios. This is a composition and institutional-access argument derived from the defined conditioning mechanism; it is not a numerical claim about those countries without comparable data.

COVID-era and other mortality shocks

A pandemic can affect the numerator through excess deaths and the denominator through lapses, migration, claims delays, changes in insurance sales, and temporarily altered underwriting. It can also interact with occupation, health inequalities, and sums assured. The SOA's 2018–2024 data-request objectives explicitly seek to study COVID-period mortality and exposures, demonstrating the need to analyze a shock through both quantities.

A coherent shock study first chooses a baseline death-rate process, then compares actual deaths in matched risk sets, treating calendar periods explicitly. A/E against an outdated table and excess mortality relative to a counterfactual no-shock model are separate estimands. If the portfolio composition shifts during the shock, part of the apparent excess may reflect changes in who remains insured. Claim reporting delays should be analyzed before interpreting a recent mortality reduction as genuine. Cause-specific evidence requires additional coding scrutiny.

The WHO ICD-11 framework supports international mortality classification, while each experience study must document how its claims-based cause information aligns with official vital records. An insurance portfolio whose death claims are well established can still have missing or heterogeneous cause-of-death data. Aggregate insured mortality conclusions should be kept independent of uncertain cause assignment.

Contrast with annuitants and the direction of selection

A person buying a life annuity receives payments contingent on survival; a person buying death protection receives a death benefit upon an insured event. These differing payoffs create different purchase incentives. This dossier uses the contrast only to show why selection cannot be assigned a permanent sign by the word "insurance." The SOA–LIMRA 2020–2024 payout annuity study reports 3.1 million contract-years of exposure and 143,190 deaths in its publicly described population, and explicitly identifies comparison with U.S. population mortality as one of its research purposes. Its annuitant results are not transferable to death-benefit life insurance without a separate product-selection argument.

The sign of economic selection can reverse between survival-contingent and death-contingent products even when both are administered by life insurers. Moreover, annuitants differ in age, distribution, financial resources, and contract entry conditions. The comparison clarifies the mechanism; it provides no warrant to use an annuitant mortality table as a term-insurance pricing basis.

Counterevidence and identification tests

Three competing explanations can produce a lower observed insured rate: underwriting reduces the proportion of high-mortality applicants accepted; wealthier or healthier individuals disproportionately choose and sustain coverage; and policy exit alters the remaining cohort. The first is a process-dependent selection mechanism, the second a demand-and-access mechanism, and the third a longitudinal risk-set mechanism. All can operate simultaneously. Evidence that one exists is not evidence that the others are absent.

The Cawley and Philipson (1999) findings are counterevidence to overly simple adverse-selection predictions in life insurance. The Einav–Finkelstein (2011) review shows why no single risk-demand correlation is diagnostic across all insurance settings. A credible test must define price variation, coverage choice, the time order of information, observed health risks, and what the insurer already knew. A change in insurance status or amount may be driven by household needs without any new private health information.

Candidate falsification and robustness analyses include the following. Compare underwriting cohorts within narrow issue-year cells rather than pooling all years; estimate both lives- and amount-weighted measures; test whether a duration pattern persists after controlling for attained age and calendar year; compare alternative standard populations; exclude data near reporting cutoffs; assess the effect of late claims; re-estimate after removing very large policies; and examine policy lapses by health-related proxies and premium step-up dates. A finding that disappears under elementary risk-set harmonization should not be interpreted as a stable mortality selection effect.

Limits of the reviewed source corpus

Industry studies are generally based on voluntary or defined contributing companies and may contain confidential or restricted cells. U.S. company composition, British CMI membership, and Irish study inclusion vary. Some underlying data are available only under subscription or confidentiality conditions. The monograph examines public institutional documents, selected methods protocols and professional standards, and publisher abstracts of academic articles. It does not claim access to proprietary microdata, full restricted CMI exhibits, or full-text copies of every paywalled academic article. Accordingly, its original mathematical illustrations and interpretive comparisons are independently reproducible, while empirical results are attributed only at their verifiable public level.

Scientific Synthesis, Applications, and Research Agenda

A consistent hierarchy of questions

The comparative evidence supports a hierarchical research practice. First identify the measured outcome and exposure unit. Next define the population whose mortality is being estimated. Then identify the process by which lives entered and remained in that population. Only after these steps should a table be selected, a mortality difference calculated, or a selection hypothesis proposed. Inverting this order—starting from an observed ratio and searching for an underwriting explanation—invites confounding by demographic composition, product design, and time.

The following five questions provide a practical analytical sequence. What was the exact covered risk set? What was the event and censoring rule? Which time axes and risk attributes were aligned? What benchmark and weighting scheme generated the comparison? Which remaining associations can be interpreted descriptively, and what additional evidence would support a causal claim? These questions establish a reproducible comparison protocol, not a new theoretical framework. They organize established survival-analysis and actuarial reasoning for the specific object of insured-life mortality.

For a prospective insured portfolio, the central scientific quantity is a conditional distribution of death given entry, underwriting, contract, survival and continued coverage. A national mortality quantity refers to a separately defined resident population. The two quantities may be compared descriptively after standardization. Direct attribution of their difference to an intervention requires a credible design addressing unobserved health, insurance demand, socioeconomic access, informative policy exit, and temporal change.

Decomposing descriptive mortality differences

Consider insured and population central mortality rates standardized to the same target-age weights. Their difference can be represented schematically as a difference in conditional rates across a set of observed strata plus an unresolved within-stratum difference. When a study uses a crude rather than standardized contrast, it also incorporates the change in strata weights. Thus the observed crude mortality difference combines at least composition and within-cell rate variation. The latter in turn can reflect remaining covariates, different mortality processes, and measurement. This is an accounting statement about rates, rather than a uniquely identifiable mechanistic decomposition.

A more granular analysis can calculate sequentially adjusted results: first age-sex, then calendar-period, then product and risk-class, then issue-cohort and duration, then socioeconomic variables where legitimately available. Each adjustment changes the question answered, so a table should show both the target estimand and variables used rather than announcing a single "corrected insured mortality rate." Over-adjustment can also obscure the total selection mechanism: adjusting for a variable partly determined by underwriting or policy persistence may remove part of the very process under investigation.

In causal analysis, conditioning on issued insurance can introduce selection bias because health, income, applicant behavior and underwriting together determine inclusion. Conditioning on surviving and remaining in force can add further bias. This is why even excellent prospective experience data do not automatically identify the effect of underwriting decisions on mortality. The measured mortality of accepted lives is simultaneously a function of individual risk and selection into observation. The evidence from Cawley and Philipson (1999) and the broader Einav–Finkelstein insurance-market review illustrates the value of testing risk-demand relations rather than assuming their sign.

Minimum reproducible insured-mortality study design

A reproducible insurer or intercompany study should specify the covered insurer set and observation years, inclusion and exclusion criteria, the product family, policy status, issue dates and age calculations, the underwriting classification as originally issued, all subsequent classification and benefit changes, and whether the measured exposure is per person, per policy, or per amount insured. It should document death definitions, late claims, overlapping policies, data-closure dates, reinstatements, conversions, terminations and reinsurance treatment. It should provide a transparent mapping from administrative records to the denominator and the mortality event. The ASOP No. 23 data-quality standard offers a relevant U.S. professional reference for the responsibility to recognize data limitations and their potential material effects.

For scientific replication, the study should additionally preserve deidentified record dictionaries, extraction and transformation specifications, aggregation rules, model version, table version, adjustments, and uncertainty method, subject to privacy and contractual constraints. A third party should be able to recreate published tables from permitted data without guessing how partial years, multiple contracts, or deaths after termination were handled. Exact personal identifiers need not enter public research packages. Aggregate provenance can be sufficiently detailed for independent audit while remaining privacy-protective.

Industry pooling improves precision but creates a new comparability problem if contributors have different risk composition and data practices. A study should therefore report company participation by period, the proportion of exposure in each major risk class, and sensitivity to influential contributors where disclosure permits. The historical construction described in the 2015 VBT methodology illustrates why the sampled company and policy populations are part of the definition of a published table.

Falsifiable duration hypotheses

The broad hypothesis that recently underwritten lives experience lower mortality than comparable longer-duration lives can be tested only after holding attained age and risk class sufficiently constant, or after explicitly modeling those differences. A select advantage may weaken with time, but the observed relationship can also depend on lapse and survivor selection. Its direction and duration are empirical properties of the studied portfolio, not universal natural constants.

A sound test would construct issue-year cohorts and estimate conditional hazards across duration, attained age and calendar time, accounting for underwriting class and benefit amount. It would then compare models with different functional descriptions of duration dependence, assess out-of-sample calibration, and determine whether the result remains under alternative policies for conversion and lapse censoring. If health information is refreshed through reunderwriting, both original and refreshed assessment dates should enter the design.

For duration-specific exits, the study should report deaths, lapses and contract transitions separately. Under a competing-risks construction, each in-force death rate is conditional on remaining covered. Missing post-lapse mortality is a limitation on conclusions about policyholders as persons, even when in-force claims are perfectly observed. The post-level term lapse and mortality study shows that the premium schedule can be associated with concentrated lapse and mortality experience, making duration-only explanations insufficient.

Risk classification, fairness and temporal robustness

A classification may predict aggregate losses yet change its meaning when the applicants, medical technology, collection rules or legal permissions change. A class named "preferred" in one insurer or period need not represent the same risk distribution in another. The ASOP No. 12 classification standard is useful for separating expected-loss relevance, objective measurement, practical implementation and constraints of law. International comparisons additionally require local legal analysis; a U.S. standard is not a global permission rule.

A useful governance test asks whether a reported mortality difference remains after the intended target group and adjustment scheme are made explicit. A second asks whether the advantage persists through different calendar shocks, including pandemic intervals. A third asks whether the relationship holds separately for policy count and amount, small and large sums assured, and major underwriting cohorts. Results that vary meaningfully across these definitions should be reported as heterogeneous rather than collapsed into one apparently definitive figure.

The CMI 2022–2024 investigation and its correction notice provide a concrete reminder that even authoritative industry results need version-specific scientific provenance. When an institutional working paper corrects data displays, secondary analysis must identify the corrected release and preserve the distinction between public summaries and restricted detailed exhibits.

International research priorities

Cross-country actuarial evidence would advance substantially through a documented common data dictionary for policy issue, medical underwriting, smoking categories, insured amount, occupation, contract status, lapse, death and exposure. Such a dictionary should support differing regulatory disclosure regimes and retain national definitions rather than erasing them. International comparisons should distinguish insurer-administrative data from vital registration and demographic estimates as explained in the Human Mortality Database protocol and UN World Population Prospects documentation.

A priority for empirical research is the joint measurement of entry and exit selection in individual life insurance. Application and underwriting records reveal who can enter; longitudinal contract histories reveal who remains; mortality linkage, when lawful and feasible, could clarify outcomes after termination. A second priority is the effect of new underwriting modalities on the interpretation of historical preferred and standard classes. A third concerns socioeconomic transportability: industry studies should show how far observed insured cohorts overlap with population reference groups before generalizing mortality differences.

Future studies should preregister their primary mortality contrast, preserve alternative benchmark results, and distinguish exploratory subgroup findings from confirmatory tests. When administrative change or a policy reform plausibly supplies exogenous variation, researchers can examine causal effects under explicit identification assumptions. Otherwise, source-faithful descriptive findings remain scientifically valuable without an inflated causal claim.

Implications for actuarial decisions

For product pricing, an insured mortality estimate should reflect the prospectively expected applicant pool, underwriting method, product design, coverage amounts, and persistency. Historical A/E is evidence for an assumption review, with adjustments for credible changes in the portfolio and external mortality. An observed favorable ratio does not justify applying the same selection discount across every product or future issue cohort.

For reserving and solvency work, conditional future deaths depend on the composition of the in-force population and its transitions. The rate at which older underwritten groups become ultimate cannot be inferred independently of lapses and benefit changes. The meaning of a table's ultimate segment is therefore actuarial and evidentiary; it may be suitable for valuation under its specifications without describing a biologically unselected population. The ASOP No. 25 credibility guidance supports the need to consider the relevance of pooled experience when local cells are sparse.

For research communication, every reported mortality percentage should state whether it is a crude rate, a standardized rate, a table-relative A/E, or a modeled excess-mortality ratio. It should identify whether deaths and exposure refer to policies, distinct lives or insured amounts. It should distinguish observed differences, projected assumptions and causal conclusions. This is the minimum evidentiary discipline necessary for the same numbers to remain intelligible across underwriting, actuarial science, demography and public risk research.

Conclusion

Insured-life mortality experience is a conditional statistic generated by the interaction of individual mortality risk and the processes of obtaining, classifying, maintaining and terminating life insurance. At the time of policy issue, the accepted cohort is selected from the population through application, eligibility and underwriting. As time passes, deaths, lapses, premium changes, conversions and other contract transitions continually change the risk set. Consequently, the relationship between insured mortality and general population mortality is time-dependent, portfolio-specific and sensitive to measurement conventions.

The international evidence establishes the institutional reality of these distinctions. The U.S. industry studies supply large selected-life datasets and valuation reference tables with specified coverage restrictions. British and Irish CMI work provides explicit experience by assurance class, period and weighting basis. Population demographic resources define a different observational universe. Academic insurance-market studies show that the relationship between risk and demand cannot be assigned a universal adverse-selection sign. Frailty research explains an additional survival-selection process operating even with unchanged individual hazards. These sources are complementary because they investigate different components of insured-life mortality, and their findings are transferable only under documented comparability assumptions.

The formal comparisons in this monograph demonstrate four central results. First, crude insured-to-population mortality differences can be generated entirely by differing risk composition. Second, standardization clarifies the comparison but leaves potentially important unobserved differences and does not identify causal underwriting effects. Third, a duration-related change in mortality can reflect underwriting, survivor frailty, informative policy exit and changing issue cohorts simultaneously. Fourth, policy-count and amount-weighted experience answer different contractual and financial questions, each requiring a correctly matched numerator and denominator.

The scientific contribution is an evidence-grounded integration of these established processes into explicit comparison and validation requirements for insured-life mortality research. Stronger empirical conclusions require longitudinal, appropriately linked and well-documented policy-level observations, consistent reference definitions, sufficient overlap between populations, and research designs matched to the hypothesis. The resulting standard of interpretation is direct: report what population was observed, how it was selected, how exposure was counted, what comparison was made, how uncertainty was assessed, and which causal questions remain unresolved.

The work is a completed internal research manuscript. Its publication date and public scientific identifier remain unset until an independently authorized release and production verification.

References