The Measure of the Wording

Abstract

Serious-illness cover pays a lump sum when the policyholder suffers a contractually defined condition; whether a given heart attack, stroke or cancer is paid turns on the written definition, not the diagnosis alone. This paper measures those definitions. For each of the five domestic offices, the current definition of each of the three families that carry the product's value — invasive cancer, heart attack, stroke — is read as written against the corresponding Irish event population and converted into a measured coverage coefficient: the share of the medically defined event population the written gate captures, reported as an interval under stated conventions, with survival-period clauses as a separate layer. Four offices capture effectively the full admitted heart-attack and stroke populations. One office's heart-attack definition — typical symptoms, new electrocardiographic changes and troponin at fixed numeric thresholds — fails 34 to 53 per cent of confirmed events as written, and its permanence-only stroke definition fails 20 per cent of stroke events at base; yet the same office carries the market's lowest prostate tumour-stage bar, giving it the market's highest measured cancer coverage. Assembled on the market-common event-value basis, the offices' blended three-family positions at age 40 are 100.9 to 101.6 against a mean of 100 — except that office, at 94.7 (scenario interval 88.1 to 95.9). The differential is concentrated in men: 93.3 to 95.2 across entry ages against 95.8 to 96.3 for women, a wording spread of roughly eight points for men against five for women. Every result is an interval over named, sourced parameters; the workbook, its checks and its correction register are the computational record.

1. Introduction

A serious-illness policy is a promise defined by its wording: the condition list names what is covered, and each named condition's definition decides which real events qualify. This paper measures the second of those, for the three conditions that carry most of the product's value, across the five domestic offices. Within the series: MWP-2026-05 computed a fair-value index across the five offices and MWP-2026-06 decomposed the cover's value, both conditioning on a single-source ordinal ranking of the wordings — a limitation each disclosed. This paper replaces that ranking, for the three families that carry the value, with measurement; and the measurement's most consequential finding is about the series itself. The ordinal scores were not only a ranking but the earlier engine's coverage coefficients, and substituting the measured coefficients collapses the -05 headline — fair value no longer varies "several times more than premium" but by roughly the premium's own dispersion — changes the -05 ordering, and strengthens -06's concentration floors. Those corrections are issued with this paper (MWP-2026-05 Erratum v1.4.0 and revision v2.0.0; MWP-2026-06 corrections note), and §4.6 carries the bridge. Measurement, in other words, narrows the market where the proxy widened it — and relocates the real differences to the specific definitions this paper prices.

The question is the prospective one a purchaser faces: given the definitions in force today, what fraction of the medically defined event population does each office's wording actually pay on? The question is answerable from public evidence because each family's definitional gates are expressed in clinical classifications — tumour stage and grade, cardiac biomarkers, neurological outcome — whose Irish population distributions are published. Where a gate excludes a slice of the event population, the slice can be counted; where the text is ambiguous, the ambiguity is construed under a stated convention and both ends reported. Claims experience constructs no coefficient: a coefficient built from claims would embed underwriting selection, lapse behaviour and notification practice, which are not properties of the words. Published claims experience is used twice, directionally and in both directions, as §4.5 reports.

The basis of measurement is contractual entitlement: what the written definition, read as a contract, covers. That basis is defensible on its own terms, and it is the basis on which definition-based declines are regularly upheld in the Financial Services and Pensions Ombudsman's legally binding decisions, where "did not meet policy definition of illness" is a standing category. It is not, however, the same as adjudicated outcome. The FSPO's jurisdiction is hybrid: conduct lawful under the contract may still be found unreasonable or unfair, and the courts have confirmed both the breadth of that jurisdiction and the deference owed to it on appeal. The two High Court decisions this series previously cited for a stronger proposition do not bear it: Geoghegan v Financial Services Ombudsman [2015] IEHC 217 concerned a condition not on the specified list and was decided on the deferential serious-and-significant-error standard of appeal, and O'Regan v Financial Services Ombudsman [2015] IEHC 85 concerned whether serious-illness cover existed under the policy at all. The United Kingdom evidence runs both ways: ombudsman decisions have upheld declines against the older troponin thresholds, while actuarial product literature records that claims just under a threshold are often paid in practice. This paper therefore presents its coefficients as measurements of contractual entitlement, reports observed claims practice beside them where the offices' disclosures speak, and claims no more for either. (The correction of the legal frame is recorded in the erratum.)

The paper's scope is the three families; everything else — benefit amounts, the partial-payment catalogues, the bounded tail of MWP-2026-06 — is held at its prior measurement, so that the wording layer is isolated. §2 summarises data and method, §3 the design, §4 the results — of which §4.4, the decomposition by sex, is the principal exhibit — and §5 the discussion. Construction conventions, per-family construction notes and references are appended; the source and correction registers are internal, and the erratum and correction record are available to interested parties on request for bona-fide verification.

2. Data and method (summary)

Definitions. Each office's operative definitions were captured verbatim from the offices' current published policy conditions and definition guides, including dated issued or current-edition contract booklets read in full for two offices; the prevailing-edition rule applies throughout. Survival-period clauses were read from the same documents. All definition captures were made between 20 and 30 October 2026, and "current" throughout this paper means the prevailing edition as at that capture window. [Clause_grid_08]

Event populations. The common event basis is ages 30 to 64. Cancer: National Cancer Registry invasive registrations 2017–19 (28,704 in-band events; male share of in-band events 47.55 per cent, the sex weight for blended coefficients), with site, stage and grade distributions from the Registry's reports and the national prostate-cancer registry cohort. Heart attack: hospital in-patient admissions for acute myocardial infarction (I21–I22) by sex and age band, 2017–19, calibrated from episodes to persons: hospital discharges double-count transferred patients and include subsequent and type-2 infarctions, so the heart-attack legs carry a person/discharge factor of 0.934 — the national audit's person estimate of approximately 6,000 heart attacks per year against 6,423 discharges per year — with a floor of 0.795 from the audit's own measured person-to-episode ratio for STEMI at PCI centres, a deliberately adverse bound. Three weaknesses of the person calibration are stated: the person estimate and the discharge counts are drawn from different years; the person estimate is a round national figure that may itself rest partly on the same hospital data, an undisclosed potential circularity; and it is an all-ages figure applied to working-age legs, where transfer duplication is more intense — the floor is carried against all three. Stroke: admissions coded I60–I69 — the interim basis is I60–I69 as coded — multiplied by an age-specific acute-stroke factor of 0.697 (interval 0.645 to 0.749): the national stroke audit's under-65 acute share applied to its acute total, against the working-age hospital count. The outcome layer is the audit's discharge distribution on the modified Rankin Scale — mild disability 34 per cent, moderate-to-severe 35 per cent, in-hospital death 11 per cent, and a no-symptom remainder of approximately 20 per cent that is a derived residual of the published categories, not a printed figure. Age-band structure: the hospital rates are published in broad bands (15–44, 45–64); diagnosis-specific five-year tabulations are not published, and the sensitivity of every age-shape statement to that banding is measured with a fitted age curve in §4.4. Denominators are Central Statistics Office population estimates. [q_inputs; q_splits_08; w_build_08 rows 47–58]

The coefficient. For each office and family, the coverage coefficient w is the share of the in-band event population, as medically defined in the national data, that the office's written gate captures, constructed per sex and reported per sex and blended. Each cell carries a low/base/high interval under the conventions of Appendix A. The survival-period layer is measured separately (§4.2) and enters multiplicatively. [Measured_w_08; w_build_08]

Verification. Every build recalculates in excess of ten thousand formulas with zero tolerated errors and re-runs the check suite: six regression tie-outs confirming the MWP-2026-05 and -06 results are recomputed unchanged on their own basis; the assembly's identity, bounds and reconciliation checks; coverage-component monotonicity across the scenario tiers for every office and family — the earlier tier-EPV monotonicity checks were retired and restated to the coverage components when the event-mix bands were folded into the tiers, because expected value is not tier-monotone once event volumes vary with the scenario; two independent-reconstruction checks that rebuild assembly cells through fresh formula chains and tie exactly; and the per-pass check blocks of the revision. All pass at the build this paper reports. [Head_EPV_08 checks; Bridge_08]

Two scope and verification notes. Stand-alone cancer-only products are outside scope; the comparison is of the five offices' full serious-illness definitions. The verification suite distinguishes regression tests — cells that confirm earlier papers' results are recomputed unchanged on their own basis (the -05/-06 lineage is computationally untouched; its correction is carried in the companion instruments, not by altering its engine) — from the reconciliation and bounds checks that police this paper's own construction. The definition texts of offices A and C were read in full from current-edition contract documents (A: the current policy conditions; C: the contract document itself, supplied by the office on request and not published); office E's three family definitions were verified against two current primary policy documents (its mortgage and protection products, May 2025 editions), which also establish that its specified-illness appendix is uniform across that product family; office D's captures match its current customer booklet in the definitions priced here; office B's clause was verified against its current definitions documentation. Every office's priced definitions therefore stand on the dated editions recorded in the internal source register, which is available to interested parties on request for bona-fide verification. Product scope is stated for each office: the definitions priced are those of the office's current standard serious-illness terms as issued under the named source documents, and any divergence across an office's product range identified under the standing correction policy is corrected in a versioned revision.

3. Design

The design is a measured head with the remainder held fixed. The three families' coefficients are constructed gate by gate from the definition texts and the population distributions; no ordinal ranking enters. The assembly applies the coefficients and the survival-period multipliers to the market-common per-heading, per-sex, per-age expected-claim values of MWP-2026-06, and normalises each age column to the five-office mean, separately and blended by sex. [Head_EPV_08]

Four design choices bear stating. First, the adverse-construction convention: an ambiguous gate reads wide, a numeric threshold reads as written. Second, the biopsy-upgrading channel is included in every differential: a tumour graded low at biopsy and upgraded at prostatectomy qualifies at every office through its Gleason limb, which recovers part of every office's failure set and shrinks the between-office differential by the same factor — at the sourced 25 per cent prostatectomy share and 67.3 per cent upgrade rate, by 16.8 per cent. Third, the scenario tiers are coherent: each tier applies one joint scenario of the underlying conditional distributions to all offices together, and the differentials the paper interprets are additionally bounded directly. Fourth, the survival layer is referenced to the shortest period in the market (ten days); common mass cancels.

4. Results

4.1 The measured coefficients (Measured_w_08; w_build_08). Low/base/high per office and family, sex-blended at the event weights; cancer additionally per sex below.

Heart attack. Four offices gate on the contemporary clinical definition — a characteristic biomarker rise with electrocardiographic or imaging evidence — capturing effectively the full admitted population; the residual 3 per cent at the low end is a stipulated evidential-friction reading. Office B's definition requires typical clinical symptoms, new electrocardiographic changes and troponin at fixed numeric thresholds ("Troponin T > 1.0 ng/ml; AccuTnI > 0.5 ng/ml or equivalent threshold with other Troponin I methods"); read against the Irish presentation mix and published troponin distributions, the threshold stack fails 34 to 53 per cent of confirmed events as written. The clause's lineage is stated: the wording is the Association of British Insurers' 2006/2011 model wording, whose thresholds were troponin-T above 1.0 ng/ml and AccuTnI above 0.5 ng/ml. The ABI's December 2014 Statement lowered the troponin-T bar to 200 ng/L and set troponin-I at 500 ng/L, aligning the definition to clinical practice (adoption by the end of December 2015); the 2018 Guide kept those thresholds (adoption by 1 February 2019); and the September 2022 Guide added diagnostic imaging as an alternative to electrocardiographic evidence and replaced the other-acute-coronary-syndromes exclusion with a myocardial-injury exclusion (adoption by 16 September 2023), clarified in April 2023. The finding is therefore about definition vintage: on the troponin-T limb the clause carries a bar five times the 200 ng/L level the UK model has set since 2014 — on the troponin-I limb its 0.5 ng/ml AccuTnI bar is nominally the model's own 500 ng/L, so the five-times statement applies to the T limb only — and its electrocardiographic requirement is one the model widened in 2022; a UK ombudsman decision upheld a decline against these older thresholds. The observed practice band is published beside the as-written band: the office's own claims disclosures bound the observed submitted-claim failure at 13 to 22 per cent (practice coverage 0.78 to 0.87) against the as-written 34 to 53; the gap is reported as the practice-versus-entitlement distinction of §1, with the caveat that paid-rate evidence cannot see claims deterred before submission. The as-written low end equals the troponin band's top because the symptoms and ECG conjuncts are taken as nested within the troponin failures — a named assumption resting on positive correlation (silent and electrically silent infarctions tend to be smaller); under no correlation the union of the conjuncts would run 49 to 65 per cent and the office's adverse-end position roughly one to two points lower, as the workbook's sensitivity grid shows.

Stroke. The market gate is disjunctive — permanent neurological deficit, or imaging-confirmed infarct or haemorrhage with symptoms persisting twenty-four hours — and captures essentially the full event population; two offices' gates are conjunctive (scan and twenty-four-hour deficit), which the grid records, with one office's specialist-confirmation conjunct read as post-hoc satisfiable (a named judgment). Office B pays only on permanent neurological deficit with persisting clinical symptoms. The audit's discharge distribution bounds the per-survivor failure between the fully recovered alone (22.5 per cent of survivors) and the recovered-plus-mild class (60.7 per cent); applied to survivors only, the base deduction is 20 per cent of stroke events (coverage 0.80) and the adverse end 54 per cent (0.46). Three directions are disclosed rather than netted: the distribution is at discharge, not ninety days; it is all-ages, where working-age survivors recover more completely — both push the true failure share higher at both ends — and the no-symptom share is a derived residual. The audit's definitional mix (tissue-based at a minority of hospitals) overstates every disjunctive office's coverage in level, not differentially. No stroke analogue of the heart-attack practice band exists in the offices' disclosures; the stroke coefficients therefore stand on the as-written basis alone, without a practice-layer comparator. The audit's mixed definitional basis cuts against office B specifically: an imaging-confirmed event whose symptoms resolve within twenty-four hours is already inside B's failure set but is counted as paid at the disjunctive offices, so the mix inflates their coefficients relative to B's, and correcting it would narrow B's stroke differential — a disclosed narrowing direction, with the bound awaiting the audit's sub-24-hour resolution share.

Cancer. All five offices share the modern core, and the differentials live in site gates. For prostate cancer, four offices require Gleason 7 or clinical stage cT2b and above; office B's bar is Gleason above 6 or stage T2N0M0 and above — the lowest prostate tumour-stage bar in the market. The construction: the conditional probability of low clinical stage given low grade is bounded on one population — the under-65 cohort whose 45 per cent low-grade share the construction uses — giving a consistent Fréchet ceiling of 0.45; the independence reading of 0.20 is an assumption, not a bound, and is carried in sensitivity; and the upgrade channel shrinks every failure set and the differential itself by 16.8 per cent. The B-versus-others differential is 16.8 to 20.6 per cent of prostate events. Office A alone adds a pathological limb paying on pT2N0M0 after prostatectomy; it applies to the concordant minority whose grade is confirmed rather than upgraded at surgery — nil at base, up to 7.4 per cent of A's prostate failure set at the ceiling — and A alone gates neuroendocrine tumours below WHO grade 2 and low-risk GIST out of full payment, a working-age exclusion priced here at 0.3 to 0.6 per cent of cancer events. Two blind spots of the denominator are stated in both directions: conditions A, C, D and E expressly include but B's definition lacks (essential thrombocythaemia, polycythaemia rubra vera, primary myelofibrosis, Merkel-cell carcinoma, pseudomyxoma) sit largely outside the C00–C97-excluding-C44 denominator, so B's narrowness there is invisible to this measure — and so, symmetrically, is A's and E's broader skin-cancer cover. One office (E) alone gates urothelial tumours at stage T2 and above, a real in-model deduction of 0.74 per cent blended. The uniform thyroid gate enters per sex (0.60 per cent of male, 1.81 per cent of female cancer events, scaled by the staging bracket).

4.2 The survival-period layer. Office A requires ten days' survival on the stand-alone basis; the other offices fourteen days on the bases examined. The measured differential is the death-window between day ten and day fourteen: 1 to 2 per cent of stroke events — early stroke mortality carries a second, immobility-driven peak that the window straddles — against 0.1 to 0.7 per cent of myocardial infarctions, whose early mortality is front-loaded; the cancer window is 0.1 to 0.7 per cent (base 0.3), built from registry-linked working-age legs. Varying the survival-window parameters alone between their sourced ends moves the blended age-40 position of the most affected office by 0.13 index points (basis: blended index at entry 40, all other parameters at base).

4.3 The assembly (Head_EPV_08). Applying the coefficients and survival multipliers to the market-common event values and normalising each age column to the five-office mean of 100, the blended base positions at age 40 are: A 101.6, B 94.7, C 101.4, D 101.4, E 100.9; across entry ages 30 to 50, B runs 94.4 to 95.7 and the four other offices 100.6 to 101.7. Under the scenario tiers — which carry one joint scenario per tier including office B's favourable low-stage reading at the offices-favourable end (0.25, a named independence-leaning assumption) and the event-mix calibration bands paired with each scenario — B's age-40 interval is 88.1 to 95.9 around a base of 94.7. The tier outputs are reported as what they are, scenario results under the named pairings, not formal bounds: B minus the mean runs −11.9 / −5.3 / −4.1 across the three scenarios, and A minus the conjunctive pair +0.47 / +0.13 / +0.28, the last figure conditional on the neuroendocrine/GIST band as §4.4 states. The compression is the first finding: an office whose heart-attack definition fails a third of events as written and whose stroke definition fails a fifth of events at base nets out about five index points below the market — because cancer carries the majority of three-family event value at every age examined and that office's measured cancer coverage is the market's highest. Clause severity and book-level cost are different quantities.

4.4 Decomposition by sex (Head_EPV_08). The blended figure averages over two materially different purchasers. The family mix of three-family event value at base runs:

Normalising within sex, the base positions across ages 30/35/40/45/50 are:

For a man, office B sits 6.3 to 6.7 index points below the five-office mean at entry ages 30 to 45, narrowing to 4.8 at 50 (a separate quantity — an office's own wording cost against hypothetical full coverage — is 5.1 to 5.3 per cent of value for the model-definition offices); for a woman, 3.7 to 4.2 points, and office B carries the market's highest measured coverage on the family that dominates her event value. The wording spread is roughly eight points for men at entry ages 30 to 45 (6.3 at 50) against five for women. No age-shape claim is made: a fitted-age-curve sensitivity, anchored exactly to the published band totals, shows the within-male age pattern — a dip at 45 and a rebound at 50 — to be an artefact of the two-band hospital rates: under fitted five-year rates the male heart-attack share is 25.1 to 25.4 per cent at every entry age and B's blended spread across entry ages collapses from 1.3 points to 0.15. The flat-band figures are the operative, sourced numbers; the age profile within them is not interpretable and is not interpreted. Office A's margin is dual-reported: with the 0.3–0.6 per cent neuroendocrine/GIST band (a stipulated band, pending registry morphology pricing) A stands at 101.6 blended at 40 with a +0.13 base margin over the conjunctive pair; without it, 101.8 and +0.47. Every A-margin statement, and the cancer ordering of the §4.6 bridge, is conditional on that band; the stroke finding of the bridge is not. The ten-day survival period contributes about half a point for men on the stand-alone basis before the neuroendocrine/GIST gate nets A's margin to +0.20 for men and +0.07 for women at base; A's effect is stroke-weighted and applies to both sexes, and its credit limb is prostate — it is not a male-vascular differential in the way B's is.

On the stand-alone basis reported above, A's survival-period advantage nets as stated; on the accelerated basis the ordering differs and is reported with equal prominence: C first (101.7), A mid-pack (101.4, level with D 101.2 and E 101.1), B last (94.6), blended at 40 — the ten-day window does not operate there and A's neuroendocrine/GIST band does. Every headline sentence in this paper names its basis.

4.5 Robustness and the claims record. The findings relied upon survive their scenarios: B sits below the five-office mean in every scenario, both sexes, all examined ages (a standing workbook check); the four non-B offices sit within 1.1 points of one another at base, so no ordering among them is asserted beyond that proximity — and on the accelerated basis their ordering shifts within that band, as §4.4 reports. Against the offices' claims disclosures, the evidence now runs one way with a caveat rather than two ways (an earlier exhibit compared unlike denominators — expected three-family shares against observed all-conditions shares — and is corrected in the erratum). B's serious-illness paid rate is the lowest disclosed (87 per cent against 92.8 and 89). And on the one like-for-like figure in the public record — B's own finer split of its serious-illness claims (cancer 67, cardiac 16, stroke 6) — B's cancer share of three-family claims is at least 75.3 per cent (a floor, since the disclosed cardiac category includes procedure headings outside the heart-attack family), which matches the as-written expectation of 75.9 per cent and exceeds the practice-band prediction of 72.8 to 74.1: the claims mix, measured like-for-like, is consistent with the as-written bite operating at the submitted-claim stage. The confounders are stated — a single office, a count rather than value basis, book mix, claims never submitted, and insured-book selection, which raises the cancer share of an insured book's claims relative to any population-based prediction regardless of wording (§5) — and no cross-office mix comparison is made, the peers publishing no three-family split.

4.6 Bridge to the series (Bridge_08; the companion instruments). The ordinal layer this paper replaces was not only MWP-2026-05's ranking but its engine: the rank-quintile scores entered the -05 fair-value computation as the coverage coefficients themselves. Substituting the measured coefficients for the three principal families — all else held as -05 published — collapses the -05 result: fair-value dispersion falls from 2.4 times (men) and 3.3 times (women) to 1.1–1.2 times at most ages against premium dispersion of 1.08, the ordering changes (A leads for men at three of the five entry ages, with D ahead at 30 and E at the age-40 price step; B, previously first or second, is last at most ages), the largest remaining per-age divergence is the price step -05 itself identified, and index levels above 1 quantify its disclosed selection caveat. Those corrections are issued with this paper as MWP-2026-05 Erratum v1.4.0 and revision v2.0.0. For MWP-2026-06 the effect is the opposite in kind: its protected findings were deliberately placement-free, the measured configuration lies inside the interval over which they were proved, and its concentration floors hold and strengthen (measured three-family shares 98.4 to 99.1 per cent; heads inside its own ceiling envelope) — a corrections note updates its point values. At family level the ordinal and measured layers agree on heart attack (the same office lowest in both), reverse on stroke (the ordinal grid ranks A last where measurement puts it joint-first), and reverse conditionally on cancer (A above C and D ordinally, below them measured once the neuroendocrine/GIST band is applied; at a nil band the cancer comparison is a tie). The lesson the series draws is stated at §1: the proxy widened the market; the measurement narrows it, and locates the real differences.

5. Discussion

For a consumer or an adviser, the practical content of §4 is a change of question: the answers differ by who is asking. A man in his forties comparing the market faces a wording spread of roughly eight points, concentrated in two definitions — a numeric-threshold heart-attack gate of 2006/2011 vintage and a permanence-only stroke gate — whose contractual-entitlement severity is measured here, whose observed paid rate is the market's lowest, and whose own claims mix, on the one like-for-like figure available, is consistent with the as-written bite; a woman comparing the same market faces a spread of about five points, within which the office carrying those vascular gates holds the highest measured coverage on the family that dominates her event value. The measurement itself shows that any single blended wording adjustment, applied market-wide, averages over an eight-against-five spread and misstates most purchasers' positions in both directions. A wording coefficient is one input to the value of a policy, not a verdict on it: premium, benefit amount, the partial-payment structure and the bounded tail sit alongside it, and this paper ranks no office overall.

For the market, the findings cut in no single office's favour. The restrictive office's clause-level severity is real and long-standing, its book-level cost a fraction of its reputation, and its own claims mix, on the like-for-like figure available, is consistent with the as-written bite. The office paying from day ten holds a measurable but modest credit — stroke-weighted, both-sex, with a prostate credit limb — which its own neuroendocrine and GIST gates net to a fraction of a point. The remaining three offices are, on the three families' wording, nearly indistinguishable: the market's centre has converged on the model definitions, and wording competition lives at two offices' deliberate departures from them.

The limitations bind where stated. The event basis is population incidence at ages 30 to 64, calibrated to persons but not underwritten: insured lives carry lower and differently shaped vascular incidence (the socio-economic gradient and underwriting both bear more heavily on heart attack and stroke than on cancer), so the vascular share of an insured book's event value — and with it the restrictive office's measured distance — is, if anything, lower than measured here; the direction runs against this paper's headline and is stated, unquantified. The coefficients are dated to the recorded editions, and nothing prevents redrafting; indeed the UK standard the restrictive clause descends from has itself moved. The hospital age bands, the assay mix, the under-65 prostatectomy share and the stroke audit's definitional mix are knowable-but-unpublished or coarsely published quantities, each carried at citable bounds with its direction disclosed. Within those bounds, the conclusions of §4 are stated at the strength the public evidence supports, and no stronger.

6. Conclusion

Measured as written against the Irish event populations, the serious-illness wordings of the five domestic offices are more uniform than the market's commentary suggests and less uniform than any single number can express. Three offices are, on the three families that carry the value, nearly indistinguishable. One departs restrictively, in two vascular definitions of 2006/2011 vintage whose as-written failure intervals of 34 to 53 per cent and 20 to 54 per cent net — against the market's lowest prostate stage bar — to a book-level position about five points below the mean, and whose own claims mix, on the one like-for-like figure available, is consistent with the as-written bite. One departs generously, in a ten-day survival period and a pathological prostate limb, to a net margin, once its own site gates are priced, of +0.13 at base over the conjunctive pair (offices C and D) and about +0.32 over the mean of the three model-definition offices. The economic substance of the restrictive departure is concentrated in male purchasers, for whom the wording spread is roughly eight points against five for women. The words are worth measuring; the measurement is worth decomposing; and the measurement is worth checking against the practice it predicts — which this paper does, on the like-for-like record the public disclosures afford.

Basis of the coverage coefficients (disclosure)

Judgment is confined to the following named conventions, each with its direction: (1) the 3 per cent evidential-friction reading at the model-definition offices' low end; (2) the adverse band 0.90–1.00 on the others' prostate stage gate; (3) the independence reading (0.20) of the low-stage share of low-grade disease, resting on the positive grade-stage association, carried in sensitivity only — the Fréchet floor is zero; (4) the 25 per cent prostatectomy share as a floor, the under-65 share being higher in an undisclosed degree; (5) the pathology-credit evidence floor of zero at base; (6) the nesting of the heart-attack symptoms and ECG conjuncts within the troponin failures; (7) the post-hoc-satisfiable reading of one office's specialist-confirmation conjunct; (8) forced points at the defensible floor; (9) the one-to-one sex blend of the index (the event-weight blend moves the age-40 figures by at most 0.06 points, as the workbook exhibits); (10) the 0.3–0.6 per cent band on the neuroendocrine/GIST gate pending registry morphology pricing; (11) the event-mix calibration bands are paired with the scenario tiers as one joint scenario per tier (§4.3); within each tier the paired factors apply uniformly across the offices. Every other parameter is a documentary fact of a policy text or a published population quantity; every coefficient is a live formula over these parameters, each parameter row carrying its source.

Appendix A — Construction conventions

(i) Adverse construction. A genuine textual ambiguity is construed against the office that drafted it; a numeric threshold or express exclusion reads as written. The construction applies the same contra-proferentem principle symmetrically where it narrows a finding: the symptoms conjunct of the restrictive heart-attack gate is floored at the no-chest-pain share (atypical-chest-pain failures excluded from the floor), and "or equivalent threshold with other Troponin I methods" is read as the clause's own assay-mapping mechanism at nominal value. (ii) Intervals and coherent scenarios. Unpublished parameters enter as intervals at citable bounds; scenario tiers apply one joint scenario to all offices together, and the interpreted differentials carry direct bounds. (iii) Prevailing edition. (iv) Separation of layers. Claims experience constructs no coefficient; it is reported as validation. (v) Reference survival period.

Exclusion layout (offices C and E). The classified cancer exclusions at offices C and E — pre-malignant, non-invasive, carcinoma in situ, borderline malignancy, low malignant potential, the same five-item list office B states — are read as attaching to the cancer definition wherever the document states the list once for the specified-illness section; the clause grid records each office's list verbatim, and no exclusion is read as broader than its stated scope.

Appendix B — Construction notes by family

Heart attack (office B's threshold gate). Failure interval 34–53 per cent of confirmed in-band events: the Irish presentation mix (¼ STEMI / ¾ NSTEMI) against published peak-troponin distributions (SWEDEHEART troponin-T; the 2025 consecutive-PCI troponin-I cohort with its selection caveat; the limb asymmetry per the 2020 paired-assay study), width dominated by the unpublished hospital assay mix (the 2015 UK-and-Ireland laboratory survey anchors the troponin-T plurality). Lineage: the clause is the ABI 2006/2011 model wording; the December 2014 Statement lowered the thresholds to 200 ng/L troponin-T and 500 ng/L troponin-I, kept by the 2018 Guide; the September 2022 Guide added imaging and replaced the other-ACS exclusion with a myocardial-injury exclusion, clarified April 2023. Practice band 13–22 per cent from the office's claims disclosures, validation only. Event denominator: discharges calibrated to persons (0.934; floor 0.795, the audit-measured STEMI duplication applied to all infarction).

Stroke (office B's permanence gate). Audit discharge mRS on the shared assessed denominator; per-survivor failure 22.5–60.7 per cent applied to the 89 per cent survivor share; deductions at base 20 per cent of events. Discharge-versus-ninety-day timing, all-ages-versus-working-age recovery, and the derived-residual status of the no-symptom share are disclosed adverse directions. Event denominator: I60–I69 interim coding with the under-65 acute factor 0.697 [0.645, 0.749].

Prostate (stage and grade gates). Joint grade-stage distribution unpublished; one population (under-65, low-grade share 45 per cent); consistent Fréchet ceiling 0.45 on P(low stage | low grade) with the all-ages stage marginal disclosed; independence 0.20 as a named assumption; the others' gate at the adverse band 0.90–1.00 applied as a shared scenario; upgrade channel 16.8 per cent on every failure set and on the differential; A's pathological limb on the concordant share only; differential of record 16.8–20.6 per cent of prostate events.

Bladder, thyroid, neuroendocrine/GIST. As §4.1: the registry coding convention for the bladder denominator; the thyroid gate per sex; A's WHO-G1 neuroendocrine and low-risk GIST gates at the 0.3–0.6 per cent band, both sexes, pending morphology pricing; the inclusions and skin-cancer blind spots stated in both directions.

Survival windows. As §4.2; the isolated movement of the layer between its sourced ends is 0.13 points on the stated basis.

Appendix C — References (sources)

Provider documents are cited collectively per the series convention; the itemised per-office register is internal. Public sources:

1. National Cancer Registry Ireland, Cancer in Ireland 1994–2022 (2024); Urological Cancers in Ireland 1994–2023 (Haugh et al., 2026), including the bladder coding convention and stage trends.
2. Gordon, et al. (2024), Cohort profile: the Irish Prostate Cancer Outcomes Research (IPCOR) registry, BMJ Open 14:e090207.
3. Gordon, N., et al. (2026), Gleason Regrading After Prostatectomy — Results From the Irish Prostate Cancer Outcome Research (IPCOR) Situated in the International Context, International Journal of Urology 33(5): e70469 (doi 10.1111/iju.70469; grade-group-1 concordance 32.7 per cent).
4. Irish National Audit of Stroke (NOCA): National Reports 2020, 2022, 2024 and Summary Report 2023; Dolan et al. (2024), PLoS One, secondary analysis of INAS 2016–2019 (under-65 share and presentation timing).
5. Irish Heart Attack Audit (NOCA), National Report 2022 and 2023 (2024): person-based STEMI incidence; HIPE-versus-person coverage counts; the ~6,000 persons-per-year estimate.
6. Hospital In-Patient Enquiry (HIPE) / Healthcare Pricing Office, admissions I21–I22 and I60–I69 by sex and band, 2017–19, as compiled in the workbook.
7. Central Statistics Office, PEA11, person-year denominators.
8. SWEDEHEART registry publications (peak hs-cTnT distributions, type-1 NSTEMI and STEMI); JUPITER-26 (Diagnostics 2025;15:2390, consecutive PCI-treated MI, hs-cTnI); van Doorn et al. (2020), JALM (paired-assay comparison); McKeeman & Auld (2015), Annals of Clinical Biochemistry (UK-and-Ireland troponin platform survey).
9. OBTAIN registry in-hospital death-timing analyses (n = 7,337): In-Hospital Mortality After Acute Myocardial Infarction: Time-Dependent Risk Profile, Journal of the American College of Cardiology 2014;63(12 Suppl A) (doi 10.1016/S0735-1097(14)60146-7), and Determinants of Early and Late In-Hospital Mortality After Acute Myocardial Infarction: A Subanalysis of the OBTAIN Registry, Canadian Journal of Cardiology (PMID 36273724); O'Keefe et al. (1995) primary-angioplasty cohort; GUSTO-I as anchor.
10. Kelly et al. (2018), BMC Palliative Care 17:37 (Irish lung thirty-day mortality); Brewster et al. (2011), Br J Cancer 104 (Scottish under-65 breast/colorectal); the 2023 Clinical and Experimental Medicine German lung-cancer treated-cohort early-mortality study.
11. Association of British Insurers: Statement of Best Practice for Critical Illness Cover (2006, 2011, December 2014) and Guide to Minimum Standards for Critical Illness Cover (2018; 16 September 2022; updated 14 April 2023); RGA (2013) and SCOR (2023) commentaries quoting the model wordings.
12. Financial Services and Pensions Ombudsman: Legally Binding Decisions database and decision 2019-0154; Geoghegan v FSO [2015] IEHC 217; O'Regan v FSO [2015] IEHC 85; [2024] IEHC 422; Utmost Pan Europe v FSPO [2020] IEHC 538. UK Financial Ombudsman Service decisions DRN-3330885 and DRN3825162; Groyer & Kreiensiek (2009), IAA, Critical illness products paper.
13. Milmo-Penny, D. (2026), MWP-2026-05, MWP-2026-06; Life Insurance Claims in Ireland (MCR-2024, MCR-2025), Mylife.ie/SMP Financial.

Declaration of interests

SMP Financial, trading as Mylife.ie, holds commercial agency appointments with each of the five domestic life offices analysed in this paper and earns commission on the protection business it places with them. This is an inherent conflict of interest, disclosed so the analysis may be judged on its method. The paper's safeguards are methodological: a single framework applied uniformly; stated conventions; interval reporting with every judgment named; computation exposed in the audit workbook; validation evidence reported in both directions, including where it cuts against the paper's own findings. The offices analysed operate in a regulated market and pay the majority of the claims they receive; the paper measures structural differences in contractual wording without promoting or disparaging any office. Offices are lettered A–E for presentation; because the workbook's inputs are the offices' own published documents and one office's heart-attack clause is quoted verbatim, the lettering is a presentational convention rather than an effective anonymisation, and the offices — office B in particular — are identifiable in substance from this paper alone.

Corrections policy

It is, and has always been, the policy of this series to accept feedback openly from every office examined in its papers, and to make any resulting correction in a versioned revision under the series erratum, which — with the full correction record — is available to interested parties on request. No office has been asked for, or given, pre-publication review of this paper or of the wordings attributed to it; the quoted definitions stand on the dated editions named in the source register, and any office that identifies an error in the wording attributed to it will have the correction made, credited, and published on the same basis as every other correction in the series.

About the author

Donal Milmo-Penny, QFA FLIA, is Research Lead at Mylife.ie. He has more than twenty years of experience in Irish financial services, protection and client advisory work. He is a Qualified Financial Adviser (QFA) and a Fellow of the Life Insurance Association (FLIA), and is a former Chairman of PIBA and Director of Brokers Ireland.

About the Mylife.ie Working Paper Series

The Mylife.ie Working Paper Series publishes original analysis of the Irish life assurance, mortgage-protection and serious-illness markets, to inform consumers, policymakers and the wider research community. It is published by SMP Financial Ltd, trading as Mylife.ie, regulated by the Central Bank of Ireland (C42382). Papers are made available without restriction and do not constitute regulated financial advice.

Use of AI

This paper was prepared with the assistance of an AI system, used under the author's direction for drafting, structuring and computational support. It was not used to generate data: the definition texts, population and claims figures derive entirely from the sources listed in Appendix C, and every sourced figure has been checked against its primary source. All quantitative results are computed in the accompanying audit workbook and are reproducible from the disclosed inputs; the workbook, not the prose, is the computational record. This paper was reviewed by an external referee, and the version published here carries the revisions made in response. The author retains full responsibility for the method, the analysis, the findings and any errors.

Copyright and disclaimer

© 2026 SMP Financial Ltd t/a Mylife.ie. All rights reserved. This working paper is published for information only and does not constitute regulated financial advice. SMP Financial Ltd, trading as Mylife.ie, is regulated by the Central Bank of Ireland (C42382). Registered office: 55 Ailesbury Road, Dublin 4, D04 F8C0, Ireland. CRO 315830.