The delib Library

Warning

delib is in its draft stage, and its contents are subject to change as development continues.

Overview

The delib library packages ten reference liability cash flow projection models for the individual life, pension and biometric-risk products sold in Germany, built with modelx, and, for each one, the product specification and technical notes the model was built from.

The coverage follows the German market rather than a template, and the shape it follows is the Drei-Schichten-Modell the Alterseinkünftegesetz imposed on German retirement saving in 2005. Which layer a contract sits in decides more about it than what it invests in: Schicht 1 buys a deductible contribution and a fully taxed pension in exchange for a product that may not be surrendered, commuted or assigned; Schicht 2 buys a state Zulage paid into the contract as a real cash flow in exchange for a statutory 100 % Beitragsgarantie; Schicht 3 buys the light Ertragsanteil taxation of a private annuity and imposes nothing. So basisrente and riester_rente take a slot each, and four more go to the Schicht-3 savings forms — the historic Kapitallebensversicherung, the klassische Rentenversicherung that replaced it, the fondsgebundene Rentenversicherung that dominates new business, and the Indexpolice, a German construction with no counterpart in the sister libraries. A seventh goes to the Sofortrente they pay out into.

The last three are Biometrie, and one of them is the reason a German library looks different from a French or a British one. The Berufsunfähigkeitsversicherung is the country’s flagship protection product, written against a statutory definition of disability in § 172 VVG, and it outsells plain Risikolebensversicherung in adviser attention by a wide margin. Pflegerentenversicherung completes the set, sitting on top of the soziale Pflegeversicherung rather than replacing it. Betriebliche Altersversorgung — Direktversicherung, Pensionskasse, Pensionsfonds, Unterstützungskasse and Direktzusage — is out of scope, as is Gruppenversicherung, the substitutive private Krankenversicherung, and Sterbegeldversicherung.

The models are the centre of the library. Each is a by-model-point projection of one product’s gross liability cash flows: Beiträge, Leistungen, Rückkaufswerte, charges and expenses, on the product’s own processing order and timing. None of them discounts — every model publishes the cash flows and leaves discounting, the Deckungsrückstellung and capital to a layer that consumes them.

Each one of these models reproduces a documented worked example, asserted cell by cell to the precision the notes display. The chain is deliberate and complete in both directions:

• product-spec.md specifies a representative product — a standardized composite built from publicly documented real products, not any single insurer’s contract. It records contractual mechanics, a full parameter set, the observed variation across insurers, and the rationale for every representative choice.

• technical-notes.md turns that product into a liability cash flow model on paper: model point attributes, state variables, assumption inputs, the recursions with their explicit processing order, policyholder behaviour, and a numeric worked example.

• The model implements those notes, and the library’s own tests/ assert the worked example against it. Change an assumption, and the test tells you whether the model and the notes have parted company.

• sources.md lists every source the first two cite, with URLs where they are known, access dates and retrieval status.

Every quantitative parameter in the library is either source-tagged or marked std — a standardization introduced for the reference implementation, carrying its rationale and, where available, the observed range across insurers. Facts taken from source material are never silently mixed with assumptions made to complete a model.

How much of this library has been checked against the documents it cites

delib was drafted blind and has since been re-verified. The two conditions are worth keeping apart, because a reader’s confidence in any one claim depends on which applies to it.

Drafted blind. The build environment blocked direct HTTP egress to every host outside a package-registry allowlist — gesetze-im-internet.de, bafin.de, aktuar.de, gdv.de, destatis.de, dejure.org and eur-lex.europa.eu were each tried and each refused at the gateway — and the session’s web-search budget ran out partway through the regulatory research. The first draft therefore rested on the authoring model’s own knowledge of German insurance law and practice, disciplined by std and unverified tags.

Then re-verified. The policy was lifted and every citation re-checked against the document it names. All fifteen German instruments the library cites were read as canonical XML from gesetze-im-internet, each with its amendment status (Stand) recorded, and 950 statutory section references were checked: 950 were correct. Insurer AVB, Verbraucherinformationen and Produktinformationsblätter were retrieved as PDFs and read.

Where that leaves each claim. Of 969 source entries, 613 now read Retrieved: yes, 37 were reached in part, and 319 could not be opened — 404 at the cited address, a consent or JavaScript wall, a paywall, a subscription login, or an address that was never established. So two entries in three rest on a document someone opened; the rest remain pointers rather than certificates, and say so individually.

Read a claim against its own entry. Coverage is uneven by design of the sources rather than of the library: the statutory core is near-complete, and the products whose primary sources are carrier wordings behind a document portal are thinnest.

The re-verification changed things, which is the strongest argument for reading the entries. It corrected sourcing, figures and attributions across every product — in one case retiring a sentence the drafted text had leaned on that does not exist in the document it was attributed to. Each product’s sources.md records what changed.

These are mechanics demonstrations, not pricing or reserving results

Every biometric basis shipped here is a std proxy. The tables a German insurer actually prices and reserves on — DAV 2008 T for death cover, DAV 2004 R and its Bestand variants for annuities, DAV 1997 I and DAV 1997 TI for Berufsunfähigkeit, DAV 2008 P for Pflege — are the property of the Deutsche Aktuarvereinigung, are not published openly, and are cited by name throughout this library and never redistributed. Nor is there a public rate card: German pricing is quote-driven, and what a Produktinformationsblatt must disclose is the contract’s own figures rather than a tariff. Replace both with company data before drawing any conclusion from the numbers.

The models

Model names are <product>_<country>_<grid>: the short form the German market itself uses where there is one — KLV, RLV, BU — a short descriptor where there is none, then DE, then _A for an annual step or _S for a monthly one. The grid letters follow lifelib, where annuallife/TradLife_A is the annual-step model and basiclife/BasicTerm_S and savings/CashValue_SE are the monthly ones. All ten models here are _S: every model in this library steps monthly, which is why the Grid column below reads monthly on every row. S carries a second sense in lifelib — scalar, one model point at a time, as against the vectorized _M models — and that is true of all ten here as well.

Kapitalbildende Lebensversicherung und private Rentenversicherung (Schicht 3)

Product	Model	Grid	Representative design
Kapitallebensversicherung	KLV_DE_S	monthly	The library’s Überschussbeteiligung chassis, and the place the German crediting arithmetic is settled: the declared laufende Verzinsung contains the Rechnungszins, so the interest surplus is max(0, decl_rate − rechnungszins) — 1,70 pp on a 2,70 % declaration against a 1,00 % guarantee, never 2,70 pp on top of 1,00 — and it is struck on the closing Deckungskapital, not on the sum insured and not on the premium. Three reserve constructions travel together and the customer gets the third: the gezillmerte Deckungskapital, the § 169 Abs. 3 VVG floor that spreads acquisition cost evenly over the first five contract years, and their maximum. On a long gezillmert contract the floor normally binds — 691,06 € at duration 12 on the anchor cell, which a model publishing the Zillmer reserve alone as the surrender value would take from the policyholder. All three Überschussverwendung systems ship, and the Bonussystem’s higher death benefit against the verzinsliche Ansammlung’s higher maturity is arithmetic rather than coincidence. The grid is monthly and the Überschussbeteiligung is not: the declaration, the three reserves, the § 169 value, the paid-up purchase and the Stornoabzug band all take a policy year and are bit-identical to the annual-step model’s, while the months decide which of those annual values a claim is paid — and a gezillmert contract surrendered inside its first policy year is now paid a guaranteed nil rather than the value the coming anniversary would close at, which is 992,00 € off the projected surrender outgo
Klassische Rentenversicherung	RV_DE_S	monthly	The same chassis with a conversion where the endowment has a maturity, projected to attained age 120 rather than stopped at Rentenbeginn. Two accounts, not one balance split in two: the Deckungskapital credited at the contract’s own guarantee vintage and the Ansammlungsguthaben at the declared rate, so one run carries 1,00 %, 2,75 % and 0,90 % cohorts at once — and forcing them onto a single rate misallocates between the two accounts while barely moving the total, which is why that error survives a reasonableness check on the headline. Three things happen at the Rentenbeginn instant: the Bewertungsreserven crystallise, the capital converts at max(garantierter, aktueller) Rentenfaktor — applying the guaranteed factor alone understates the anchor’s annuity by 12,5 % — and the Rentengarantiezeit begins paying on the annuitised count rather than on survivors. The grid is monthly and the deferment is not: the premium, both accounts, the § 169 Abs. 3 floor and the surrender value all take a policy year and are bit-identical to the annual-step model’s, because § 12 Abs. 1 VVG makes the Versicherungsperiode the year and § 169 Abs. 3 pays a mid-period exit the value struck at its end. What the finer grid buys is the Rente itself — a Rentenfaktor is quoted in euro a month, and the annual model paid twelve instalments at the start of each payout year, a compression worth 369,14 € of the anchor’s payout phase, all of it outside the Rentengarantiezeit where the count is fixed and the two grids agree to the cent
Fondsgebundene Rentenversicherung	FRV_DE_S	monthly	The insurer guarantees the number of Anteileinheiten and not their value, so there is no Rechnungszins, no Deckungskapital, and — the Anlagestock being a segregated § 125 VAG asset pool held in the very units the liability is denominated in — no investment-mismatch term anywhere in the model. net_cf is the non-unit stream alone; booking the whole Fondsguthaben as insurer outgo is the first-order failure mode, and the scale of it on the anchor is 64 869,36 € of benefits against 4,39 € that is actually an insurer cost. The acquisition charge is the 25 ‰ Höchstzillmersatz spread over sixty monthly instalments, and the cliff at month 61 where the Anlagebeitrag steps from 162,00 € to 192,00 € is why this model is monthly — an annual grid cannot place month 60. Two mortality bases sit in one contract and no cells reads both files: DAV 2008 T behind the Risikobeitrag on the Beitragsrückgewähr amount at risk, DAV 2004 R behind the guaranteed Rentenfaktor
Indexpolice	Index_DE_S	monthly	A conventional profit-participating contract with the capital in the Sicherungsvermögen and no unit account, unit price or fund value anywhere: the index participation is a form of Überschussverwendung under § 153 VVG with no independent statutory footing, and the declared Überschussanteilsatz is the option budget — spent on the option package or credited as interest, and allocated exactly once, which is what check_surplus_alloc() says. The payoff is a sum of monthly returns each capped and none floored, floored once at the year: the shipped path reproduces the Indexjahr in which the index rose 6,4402 % and the credit was nothing, and four plausible-looking misreadings — flooring each month, compounding the capped returns, flooring the compounded raw return, applying the Partizipationsquote to it — each print a different wrong number. The Höchststandsicherung ratchets the ledger of credits, not the balance, which falls in a year that credits nothing wherever the reserve charge sits at or above the guaranteed rate. The grid is monthly and the Indexjahr is not: the cap, the sum, the credit, the ledger and the balance all take a policy year and are bit-identical to the annual-step model’s, while the twelve capped monthly returns the Indexjahr is made of are now cells of the frame rather than a loop inside one — index_return_capped_mth(t) over a policy year sums to that year’s index_sum(k) — and the months decide which exits are deaths and which surrenders, moving 35,69 € off the anchor’s death outgo and 30,42 € onto its surrender outgo, the two unequal because a death is paid the 32 400,00 € Mindesttodesfallschutz floor and a surrender the much smaller Rückkaufswert

Geförderte Altersvorsorge (Schicht 1 und Schicht 2)

Product	Model	Grid	Representative design
Basisrente	Basis_DE_S	monthly	Schicht 1 is a list of prohibitions — nicht vererblich, nicht übertragbar, nicht beleihbar, nicht veräußerbar, nicht kapitalisierbar — and the model is one too: no cv_pp, no claims_lapse, no Kapitalwahlrecht, no surrender decrement at any duration, and § 169 VVG with its Stornoabzug simply inoperative. The absences are the product, so the test module asserts the name list, a missing cells having no formula to check. The only behavioural exit is § 165 VVG, which removes the premium and not the policy, so two ledgers run side by side carrying different account values and pols_if decrements on mortality alone: by the twenty-third policy year the in-force count has fallen only to 0,932780 while the premium-paying count is 0,512516, and the difference is a cohort still in force, still credited and still converting. Everything paid to a survivor must be paid as an annuity, so the death benefit is the reserve leaving as the single premium of an immediate annuity this model does not project, and the cover is priced through a reduction in the Rentenfaktor rather than by scaling the benefit. The grid is monthly and the contract is not: the contribution, the four charges, the declared rate, both account blocks and the § 165 freeze all take a Versicherungsjahr and are bit-identical to the annual-step model’s, while the Rente is what the finer grid buys — a Rentenfaktor is quoted in euro a month, and where the annual model booked twelve instalments on each payout year’s opening count, and named that as its own pitfall, the instalment is now paid to whoever is alive at the start of each month, which is 4 290,52 € — 1,6 % — off the anchor’s annuity outgo, and a Rentengarantiezeit is 12m instalments beginning in the month after the death that starts them
Riester-Rente	Riester_DE_S	monthly	The Zulage is a contribution, not a rebate: the ZfA pays it to the provider, it is credited, counted in the guarantee, invested and taxed at the end like any other Beitrag, and on one shipped cell the state pays 76 % of the whole contribution — so it is a published positive income column beside premiums and is never folded into it. Two different lags apply and one offset used twice reproduces neither: § 86 strikes the Mindesteigenbeitrag on the previous calendar year’s earnings while the cash arrives one projection year late, which is why the anchor’s Zulage falls between the third and fourth projection years while its premium rises — a Zulage that stops is a contribution the saver must make good. The statutory 100 % Beitragserhaltungszusage accumulates contributions without interest and is tested exactly once, at Rentenbeginn: on the low declared-rate cell the account reaches 20 481,72 € against a 21 000,00 € guarantee, and that 518,28 € Garantielücke, funded from the insurer’s own resources, is the product’s signature output — a Riester model on which it is never positive has demonstrated nothing. The grid is monthly and the subsidy is not: the Eigenbeitrag, the Zulage and its ZfA lag, the two charges, both balances, the guarantee accumulator and the conversion all take a Versicherungsjahr and are bit-identical to the annual-step model’s, the Garantielücke and the Kleinbetragsrente verdict included, while the Leibrente the AltZertG requires is now paid one instalment a month — 361,74 € off the anchor’s payout phase — and the three accumulation exits compete month by month instead of running in sequence at a year end, moving 5,62 € of death outgo and 2,64 € of surrender outgo onto 8,30 € of transfers

Biometrie und Rentenbezug

Product	Model	Grid	Representative design
Sofortrente	Sofort_DE_S	monthly	The payout chassis, and the one model in the library with no behaviour at all: § 168 Abs. 3 VVG displaces the right of termination once the Rentenbezug has begun, so there is no lapse_rate, no Rückkaufswert, no Stornoabzug and no paid-up state at any duration, and the answer rests more purely on the mortality basis than anywhere else in delib. The Rentengarantiezeit is a max and not a sum, and both errors it closes off are large and opposite: decrementing the guaranteed instalments for survival pays 7,13 % too little, adding the certain floor pays 92,87 % too much, because γ + l_a pays 1 + l_a for the whole window. The Kapitalrückgewähr makes the pricing equation implicit in the annuity and is solved, not evaluated — striking the plain annuity and subtracting a refund cost gives 318,7362 € against the correct 298,8348 €, which is not a rounding. The surface is generational and read at (attained age, birth cohort), never at (age, projection year): a period proxy overstates the annuity a given Einmalbeitrag buys by 5,1 %
Risikolebensversicherung	RLV_DE_S	monthly	The protection chassis, and where the Bruttobeitrag / Zahlbeitrag pair is derived rather than assumed: the guaranteed gross premium is struck once by first-order equivalence on tariff survivorship, and the billed premium follows from the surplus mechanic — the MindZV’s 90 % minimum allocation of the Risikoergebnis, times the tariff’s own mortality margin — reaching a ratio of 0,574725 out of the arithmetic. Setting decl_scale to zero raises the bill by 74,0 % with no change to any benefit, decrement or guaranteed term, and so with no § 163 procedure, no Treuhänder and no remedy: the largest policyholder risk in the product, and a one-Reference change. There is no cash value at any duration — § 169 Abs. 1 VVG reaches only a contract whose insured event is certain to occur — and yet a Deckungskapital builds to 7 553,29 € at duration 16 and runs off to exactly zero, so “no Sparanteil, therefore no reserve” fails the Thiele check. The § 161 three-year suicide window is a benefit switch on death claims only, applied tranche by tranche, so each Nachversicherungsgarantie increment carries its own clock. The grid is monthly and the product is not: every one of those constructions stays on the anniversary and the first-order equivalence is bit-identical to the annual-step model’s, while the finer grid puts a claim in the month it happens and collects a modal Zahlbeitrag in instalments, which is § 168 VVG’s Versicherungsperiode modelled rather than approximated
Berufsunfähigkeitsversicherung	BU_DE_S	monthly	A multi-state model with a return arc, not a decrement model: aktiv, leistungspflichtig, and a three-month run-off ledger that is § 174 VVG in arithmetic — where the insurer establishes that its liability has ceased it must still pay to the end of the third month after the notice reaches the policyholder, so a recovery does not stop the annuity in the month it happens, and the tail is 1,6 % of all benefit on the anchor cell. Death and lapse are the only exits; inception, recovery and reactivation are internal transfers, and putting them into the in-force recursion is how a multi-state model loses mass invisibly. The Beitragsbefreiung is not a benefit but the absence of a premium, and it falls out only if the premium is weighted by pols_prem rather than pols_if — the classic German BU implementation error, which leaves every total looking plausible, and the reason check_net_cf() rebuilds the premium leg from the Zahlbeitrag actually billed. Two escalations run on two clocks: the Beitragsdynamik on the policy anniversary, the Leistungsdynamik on the anniversary of each onset
Pflegerentenversicherung	Pflege_DE_S	monthly	Nine states and only two absorbing: five Pflegegrade with deterioration and Herabstufung between them, a Karenz ledger per onset, and the trigger being the statutory Pflegegrad of §§ 14, 15 SGB XI rather than a definition the insurer writes. Every grade transition is internal to pols_if, and the monthly step holds the forces constant so the competing exits share one survival probability in proportion to them — p_stay + Σ p_j = 1 exactly, which is what makes the state identity an identity rather than an approximation. Grade and mortality are correlated and the loading is stated on the force of active mortality, 1,5 at Pflegegrad 1 rising to 9,0 at grade 5, so the highest-paying state is the shortest-lived and pricing this annuity on DAV 2004 R would be prudent in exactly the wrong direction; the benefit is a grade-by-grade sum, and applying the entry-mix mean to the aggregate care population understates the whole benefit by 30 %. The Wartezeit runs from inception and gates the incidence force, the Karenzzeit runs from onset and needs its own ledger dimension — routinely conflated, and implemented in two different places

One shape, enforced

Every model has the same two Spaces — Data reads the input CSVs once per model, and Projection is parameterized by point_id — with inputs as external CSVs beside run.py, so the model folder holds formulas and nothing else. Projection’s docstring carries the mapping from the technical notes’ actuarial symbols to the cells names.

That shape is asserted rather than merely described: tests/test_model_conventions_de.py applies it to every model in the registry, and each model additionally has its own test module for its worked example and its product-specific invariants — the notes’ “Known modeling pitfalls” sections are written up there as tests.

The pairing of model name to folder is deliberately not derivable from the folder name — fondsgebundene_rentenversicherung spelled out is unusable in a model name — so it is registered once in tests/de_registry.py, and the conventions suite asserts that the registry, the directory on disk and the model’s own _name all agree, along with the country and grid tags.

The registry is per library; the contract it enforces is the one uslib is held to, and cells names come from lifelib — basiclife/BasicTerm_S first, then savings/CashValue_SE — so a name means the same thing here, in uslib, in uklib, in frlib and in lifelib. The shared vocabulary table is the settled ruling across the libraries, and the time index is the same in all of them. The time index t is 0-based: t = 0 is the first period of a policy projected from issue (the issue year on an annual grid, the issue month on a monthly one), period t runs from time t to time t + 1, and the attained age is age_at_entry + t on an annual grid (age_at_entry + duration(t), duration(t) = t // 12, on a monthly one) — and here t is always a month. proj_len() is the number of periods from t = 0, i.e. the exclusive end of the frame: result_cf() covers t = t_first, ..., proj_len() - 1, where t_first is 0 for a point projected from issue and the elapsed periods for an in-force point. This is lifelib’s own convention (basiclife/BasicTerm_S, savings/CashValue_SE: for t in range(proj_len())). A contractual policy year is the 1-based label t + 1 (duration(t) + 1 on a monthly grid) and is derived, never indexed by.

Every model steps monthly; no product is thereby a monthly product

The grid is the model’s, not the product’s. A mechanic the contract states in years stays annual and is placed in the month the contract places it, which is why five of these models — KLV_DE_S, RV_DE_S, Index_DE_S, Basis_DE_S and Riester_DE_S — run on two clocks and say so in their Projection docstring: an annual coordinate — k, the policy or projection year — for anything struck per Versicherungsjahr, and t, the month, for anything that happens on a date. A cells’ argument is the statement of which clock it is on, and the bridge between them is one cells — policy_year(t), proj_year(t). The Beitrag of a contract written annually is one instalment in month 0 of each policy year rather than a twelfth every month, because § 12 Abs. 1 VVG makes the Versicherungsperiode the year and the Beitrag is payable in advance for it; the declared laufende Verzinsung is credited once a policy year; Index_DE_S’s Wahlrecht is exercised once, at an anniversary; a § 169 Abs. 3 VVG surrender value is struck zum Schluss der laufenden Versicherungsperiode and not at the cancellation date; a § 165 VVG Beitragsfreistellung takes effect “für den Schluss der laufenden Versicherungsperiode” and so falls at an anniversary; and Riester_DE_S’s Zulage still reaches the account in the year after the contribution year it is claimed for.

What the finer grid changes is when things happen inside the year. Every annual decrement is applied twelve times at the geometric twelfth 1 − (1 − r)^(1/12) and never at r / 12, so twelve months compound back to the year’s rate exactly and the in force at every anniversary is the annual-step model’s own figure to the last bit; a rate that is a certainty rather than a probability — the terminal q = 1 that closes a mortality table — is not twelfth-rooted but placed whole in the anniversary month. That exactness is what makes the figures that do move readable as product effects rather than discretisation noise: a Rente quoted in euro a month is now paid monthly in advance (RV_DE_S −369,14 €, Basis_DE_S −4 290,52 €, Riester_DE_S −361,74 €); deaths and surrenders compete month by month, so a year’s total exits are unchanged and their split is not; a policy leaving mid-year bears administration only for the months it was there; a Rentengarantiezeit is a count of monthly instalments; and the twelve capped monthly observations of Index_DE_S’s Indexjahr — always the mechanic the product is named for, but buried inside one cells — are rows on the frame, where capped above and not floored below can be read month by month.

Six models — the five above and RLV_DE_S — publish result_cf_annual(), which regroups the same months into policy or projection years and is never a second projection, so an annual comparison is a sum of the frame that was actually run. The balances that an annual result_cf() used to carry — account values, credited interest, in-force counts — live in an annual state frame instead (result_pols(), result_acct(), result_index(), result_surplus()), so a balance cannot be summed down a cash flow statement.

The two rulings this library added

Each library in this repository settles a convention of its own and asserts it rather than describing it. delib settled two, and both are enforced by the conventions suite.

check_net_cf() is required of every model. A cash flow model’s headline number is net_cf, and until now it was the one quantity nothing checked: the roll-forward identities the models publish check policy counts and account values, and the statement that reconciles them into net_cf lived in prose. Every delib model publishes check_net_cf(), a bool over all t that reconstructs net_cf(t) from the statement’s own published parts, with the per-period residual at check_net_cf_resid(t). The identity is a product fact — a term cover reconciles premiums less claims less expenses, a unit-linked contract has to cross the unit / non-unit boundary to do it, and a payout annuity has no premium term at all — so each model.md states its own in one line and the conventions suite asserts only that the cells exists, has the CashValue_SE signature, and returns True on every model point. frlib carried the name on five of its nine models; here it is the contract.

Every assumption CSV carries a provenance column. The hard rule of all five libraries is that every quantitative parameter is either source-tagged or marked std. In the prose that rule is enforced by review; in the shipped input files it was enforced by habit, and habit is what a table added in a hurry escapes. Here it is a property of the library: each row of each input CSV carries its own tag — [S3], [REG-R21], [std] with a short rationale — and a file without a populated provenance column fails the suite. model_point_table.csv is the single exemption, because a model point is a configuration rather than an assumption: its columns are one policy’s own terms, and tagging them row by row would repeat the same fact once per policy while saying nothing about any assumption.

Given the retrieval conditions above, the second ruling earns its keep twice over. When a citation is a pointer rather than a certificate, the least a library can do is put the pointer next to the number.

What is German about these models

Five things recur across the set and are worth knowing before reading any one of them.

Every biometric basis shipped here is a std proxy. German insurers price and reserve on the tables of the Deutsche Aktuarvereinigung — DAV 2008 T for death cover, DAV 2004 R and its Bestand variants for annuities, DAV 1997 I / RI / TI for Berufsunfähigkeit, DAV 2008 P for Pflege — and those tables are the DAV’s property, are not published openly, and are cited by name throughout this library and never redistributed REG-R47 REG-R48 REG-R49 REG-R50 REG-R51. What ships beside each model is a construction, anchored so that the model’s own worked example reproduces exactly, with the anchor named in the Data docstring: mort_rate_1st(M, 37) = 0.001200 on KLV_DE_S, q_base(M, 50) = 0.002000 on RV_DE_S, qx_tariff(37) = 0.00080 on FRV_DE_S, qx(67) = 0.014000 on Basis_DE_S, ann_factor() = 20.87222879 on Riester_DE_S, inc_rate(30) = 0.001100 on BU_DE_S. Each model.md also states what a replacement must preserve rather than what it must equal — that DAV 2004 R’s surface is generational, so it is read at attained age and birth cohort and a period proxy overstates Sofort_DE_S’s annuity by 5,1 %; that disabled-lives mortality must exceed active-lives mortality state by state and never be one rate for both; that a Pflegetafel must carry incidence by grade of entry, deterioration dominating recovery above 75, and transition probabilities that sum with the stay probability to one. This is the single largest gap between these models and a production one, and it is why every model.md opens by saying the model is a mechanics demonstration rather than a pricing or reserving result.

The declared rate contains the guarantee, and the surplus it credits is a constrained allocation rather than an assumption. The German laufende Verzinsung is the Garantieverzinsung plus the laufende Zinsüberschussbeteiligung REG-R53, so the interest surplus is a subtraction — max(0, decl_rate(t) − rechnungszins()) — and never a rate paid on top of a rate. A model that credits 1,00 % and then a further 2,55 % puts 56,82 € into RV_DE_S’s first year instead of 40,82 € and reaches 63 768,69 € at Rentenbeginn against the correct 58 788,98 €, with the whole error sitting in one of the two accounts. The rate itself is an output of a constrained allocation: § 153 VVG makes participation an entitlement and the MindZV puts an arithmetic floor under it — at least 90 % of the investment result, 90 % of the Risikoergebnis and 50 % of the übriges Ergebnis REG-R18 REG-R24 — with § 139 VAG’s Sicherungsbedarf test cutting back the Bewertungsreserven share REG-R9. KLV_DE_S carries the machinery and RV_DE_S, Index_DE_S, Basis_DE_S and Riester_DE_S consume a declared rate from it, Index_DE_S spending it as an option budget instead of crediting it. The guarantee inside that rate is a cohort fact fixed at conclusion: an existing contract keeps the Höchstrechnungszins in force when it was written REG-R14 REG-R15, the rate stepped from 0,25 % to 1,00 % on 1 January 2025 — the first increase since 1994 — and a German book is therefore a layered stack of vintages rather than one rate. Four models carry the vintage on the model point (RV_DE_S runs 1,00 %, 2,75 % and 0,90 % cells in one run; Basis_DE_S runs four vintages; Index_DE_S reaches back to 0,25 %; KLV_DE_S ships both DeckRV ceilings in a cohort-keyed table and asserts them), and Sofort_DE_S prices each Einmalbeitrag against the cap of its own vintage as an inequality, because a carrier may price below it and one in the corpus is observed doing so.

The Bruttobeitrag and the Zahlbeitrag are two different numbers, and only the first is guaranteed. This is the German protection signature and it has no counterpart in the sister libraries: the contract guarantees a gross premium as the maximum the policyholder can ever be required to pay, and bills a lower net one obtained by crediting anticipated surplus in advance REG-R24 REG-R27. RLV_DE_S derives the split from the mechanic rather than assuming it — the Sicherheitszuschlag’s actuarial value at issue, times the MindZV’s 90 % minimum allocation, times the risk share of the gross premium, giving 0,574725 — and publishes prem_gross, premiums and prem_rebate as three separate columns so the gap is visible in the frame. BU_DE_S publishes the same pair as premiums and surplus_credit, holding the ratio at 0,70 and calling that its largest discretionary assumption. The consequence is a real policyholder exposure and both models make it a one-parameter stress: withdrawing the credit entirely raises RLV_DE_S’s bill by 74,0 % and BU_DE_S’s by 42,86 %, with no change to any benefit, decrement or guaranteed term, and therefore no § 163 VVG procedure, no Treuhänder and no remedy. A model carrying one premium column cannot represent the product, and a model carrying only the billed one has silently assumed the credit is permanent.

The tax layer is model structure, not a parameter. Which of the Drei Schichten a contract sits in decides what the model may contain, and delib implements the constraints as absences and mechanics rather than as flags. In Schicht 1 the five prohibitions of § 10 Abs. 1 Nr. 2 Buchst. b EStG REG-R39 mean Basis_DE_S has no surrender value at any duration, no lapse decrement, no cv_pp, no Kapitalwahlrecht and no lump sum to anyone at any date, and that everything paid to a survivor is paid as an annuity — so a death benefit there is the released reserve leaving as the single premium of a new contract rather than a payment to a beneficiary. In Schicht 2 the AltZertG’s 100 % Beitragserhaltungszusage is a nominal accumulator tested once at Rentenbeginn REG-R43, the Zulage is a real cash flow paid by the ZfA to the provider on a statutory lag REG-R42, the acquisition charge must be spread over at least five years — tighter than anything the VVG imposes on a Schicht-3 contract — and the 30 % Teilkapitalauszahlung cap and the Kleinbetragsrenten-Abfindung are computed rather than assumed. Schicht 3 imposes none of that and instead leaves a behavioural fingerprint: the twelve-year and age-62 conditions of § 20 Abs. 1 Nr. 6 EStG REG-R45 put a step in four models’ surrender tables at the duration the threshold is crossed, and FRV_DE_S keys it on age as well as duration because keying it on duration alone fires fourteen years early on its anchor cell.

Scope limits are stated rather than faked. Where a mechanic could not be established, or a deterministic run cannot reach it, the models say so instead of shipping a number. Index_DE_S carries no optimal-election rule, inertia model or within-year switching for the annual Wahlrecht, because none is established for the product family and a switching rule would put an unevidenced behavioural assumption at the centre of the result; its base run at w = 1 is declared a modelling choice made so the model demonstrates the index arm, with model point 11 shipped as the RV_DE_S comparison. FRV_DE_S implements none of the hybrid and guarantee designs — statisches and dynamisches Hybrid, Zwei- and Drei-Topf-Hybride, i-CPPI, Wertsicherungsfonds — because each is a reallocation rule along a path and a deterministic projection has one smooth path, so the rule either never triggers or triggers on a hand-chosen shock; what would have to be added is named instead. BU_DE_S ships the AU-Klausel switch on at an uplift of exactly 1,00 on one model point, because no source quantifies what six months of certified Arbeitsunfähigkeit adds to incidence and an inert switch is honest where an invented loading is not. RV_DE_S records a payout-phase administration charge and never applies it, the Rentenfaktor being exogenous and already carrying the tariff’s payout loading. And where a simplification runs one way, the direction is stated: FRV_DE_S’s omitted surplus credit biases the projected Fondsguthaben downward, its unimplemented paid-up cohort biases charge income upward, Sofort_DE_S’s independent joint lives overstate the joint-life annuity, and Pflege_DE_S’s aggregate in-care mortality understates what a Karenzzeit removes.

Chassis relationships

Products that share machinery point at the file where it is specified rather than silently restating it, and each pointer states what it inherits and where it deviates:

• KLV_DE_S is the Überschussbeteiligung chassis. The Kapitallebensversicherung technical notes are the primary home of the declared-rate arithmetic, the four-component German surplus split, the three Überschussverwendung systems, the Zillmerung and the § 169 Abs. 3 VVG floor. RV_DE_S is the same Deckungskapital and surplus machinery with a conversion where the endowment has a maturity, and carries the split by pointer rather than by restatement — only the Zinsüberschuss was established for the annuity, and inventing the other three would be inventing three rates. Basis_DE_S and Riester_DE_S add a tax wrapper and, in the second case, a state Zulage to the same accumulation; Index_DE_S keeps the chassis and spends the declared surplus on an index participation instead of crediting it, so model point 11, electing the sichere Verzinsung arm at w = 0, is an RV_DE_S comparison run inside the index model. FRV_DE_S deliberately does not sit on it: a unit-linked Rückkaufswert is a Zeitwert of fund units and not a Deckungskapital, and the contract has no Rechnungszins to declare a rate above.

• RV_DE_S is the accumulation-and-conversion chassis the two subsidised layers inherit. Basis_DE_S states its deltas against it — the same deferred annuity with the Schicht-1 prohibitions imposed on top — and Riester_DE_S points at it for the dk_pp / surplus_acct_pp recursion and for § 169 VVG, adding the Schicht-2 apparatus the classic contract has none of. Read in the other direction, Basis_DE_S and Riester_DE_S are the useful contrast with each other: a statutory Beitragsgarantie, a permitted 30 % Teilkapitalauszahlung and a Kleinbetragsrenten commutation on one, and none of the three on the other.

• RLV_DE_S is the protection chassis, and it reaches less far inside delib than a reader might expect, which is worth saying rather than leaving to be discovered. It shares the Überschussbeteiligung machinery with KLV_DE_S in a different Überschussverwendung form — surplus netted against the premium rather than credited to a reserve — and it is the file the Bruttobeitrag / Zahlbeitrag split is derived in. It does not extend to the two biometric models: BU_DE_S and Pflege_DE_S are monthly multi-state projections and share no recursion with it at all.

• BU_DE_S and Pflege_DE_S share the monthly multi-state chassis, and share it with frlib’s assurance dépendance. dis_cohorts and dep_cohorts, pols_dis_dur(t, z) against pols_pg(t, g) against pols_part / pols_tot, and pols_prem, check_states and check_pols_roll_fwd mean the same thing on all three. That is the whole of the shared vocabulary, and the two names a reader might expect beside them are worth naming as absent. cohort_len truncates a claim-duration cohort vector and so exists only where there is one — BU_DE_S and Dep_FR_S; Pflege_DE_S’s ledger is graded rather than aged and carries no such name. And the return-to-active transition is named three ways because it is not one transition: Dep_FR_S’s pols_recovery returns a life to autonomy, Pflege_DE_S’s pols_reactiv returns a Pflegegrad 1 life to pols_act paying and exposed to lapse in the same month, while BU_DE_S’s pols_recovery returns no one — it ends the benefit into a three-month run-off out of which the life reappears as pols_reactivation. Each model.md tabulates the rest of where they part: the ledger dimension differs — a claim-duration cohort in BU, a Pflegegrad in the LTC annuity, a two-level French severity in Dep_FR_S — and only the German LTC model’s is a benefit schedule. BU_DE_S’s pols_runoff_slot is the counterpart of Dep_FR_S’s pols_red: a small holding ledger a naive implementation omits, and a first-order error in both.

• FRV_DE_S shares the unit-linked chassis with frlib’s unités de compte contract, UC_FR_S, and the shared names mean the same thing on both: av_pp / av_pp_at for the fund at its named within-month timings, the unit count and its two movements, and the net amount at risk floored at zero — nar_pp here against the garantie plancher there — as the only part of a death benefit the insurer funds. Its model.md tabulates the correspondence in both directions, including the three German terms of art that keep their German form in the cells names because each names a quantity with a statutory definition: beitragssumme(), stornoabzug() and the three rentenfaktor_*().

• Across markets — the payout core. Sofort_DE_S shares the payout-annuity core with uslib’s SPIA, uklib’s pension annuity and frlib’s rente viagère: duration_mth, horizon_mths, is_payment_mth, certain_floor, payment_factor, lives_if, lives_death, annuity_pp, annuity_payments, check_lives_roll_fwd and check_payment_factor mean the same thing on all four, and result_pols() is the same second frame. Two names the other three carry are absent here and the absence is argued rather than accidental: payment_surv_mth and payment_factor_life separate the survival index of a payment from the month it falls in, and in this model the payment instant is the start of month t under both timings, so the two indices coincide and a second cells would only restate lives_if. Inside delib, Sofort_DE_S is RV_DE_S’s payout phase as a product in its own right — which is why an immediate-annuity document is direct evidence for a deferred contract’s conversion basis — it is the survivor’s single premium Basis_DE_S books and does not project, and it is the contract Riester_DE_S’s second phase now runs on the same monthly grid as.

• Across markets — the rest. KLV_DE_S copies lifelib’s annuallife/TradLife_A external-input layout, which the whole library then follows, and its nearest conceptual relative is frlib’s fonds en euros, the same idea under a different statute — with the difference that a French fonds euros credits a rate to an account balance while a German endowment credits it to a reserve. RLV_DE_S and frlib’s temporaire décès are the same product in two markets, and three differences are named so a reader does not carry one across: the French cotisation is revisable at attained age and the German Bruttobeitrag is level; the French product accelerates the death capital on PTIA and the German one has no living benefit at all; and expenses includes commission there and excludes it here. Index_DE_S’s nearest relatives are uslib’s FIA_US_S and RILA_US_S, which share the cap and participation-rate vocabulary and the annual reset but not the German financing identity: an FIA’s index budget is the insurer’s option budget on a fixed-annuity chassis, while here it is the declared Überschuss and is bounded below by the MindZV.

How to use the library

Create your own copy of the delib library, as described in the Copying a Library section. For example, to copy it to C:\path\to\your\delib:

>>> import lifelib

>>> lifelib.create("delib", r"C:\path\to\your\delib")

Each model reads from its own directory, so run one directly:

python products/klassische_rentenversicherung/run.py

or read it and take the cash flow statement:

>>> import modelx as mx

>>> model = mx.read_model("products/klassische_rentenversicherung/RV_DE_S")

>>> model.Projection[1].result_cf()

Projection takes a point_id; Projection[1] is each model’s worked-example anchor cell. result_cf() returns a tidy DataFrame indexed by t — a projection month, in every model — with one column per cash flow line. Where a model also publishes result_cf_annual(), that is the same frame summed into policy or projection years, and its annual state frame (result_pols(), result_acct(), result_index(), result_surplus(), result_fund(), result_states()) is where the balances are.

The tests ship inside the library and run against your copy:

python -m pytest tests -q

Library contents

File or folder	Description
products/<product>/	One directory per product, holding its documents and its model together. Ten of them.
products/<product>/product-spec.md	The representative product specification: mechanics, parameters, variation across insurers.
products/<product>/technical-notes.md	The liability cash flow model on paper: state variables, recursions, processing order, worked example.
products/<product>/model.md	How the model implements those notes — what was standardized, what diverges, what the tests cover.
products/<product>/sources.md	Every source the product’s documents cite, with URLs where known, access dates and retrieval status.
products/<product>/<Model>/	The modelx model itself. Formulas only — no embedded data.
products/<product>/*.csv	The model’s inputs, external to the model folder so they can be edited or swapped in place. Every assumption file carries a provenance column.
products/<product>/run.py	Reads the model and prints its cash flow statement.
references/	The cross-product regulatory and actuarial bibliography, cited as [REG-R#].
tests/	One module per model for its worked example and invariants, plus test_model_conventions_de.py for the house style, and de_registry.py carrying the model registry.
_research/	The raw research notes every citation traces back to. Provenance, not documentation — shipped but not rendered.

_research/ carries one file per product plus regulatory-actuarial.md, and records what each source is and what could be established about it. Its source lists are never renumbered: the product documents cite against them.

Citation conventions

Whether a citation tag is a link tells you what kind of source it is. [R1] and [REG-R18] are links: the first lands on entry R1 in that product’s sources.md, the second on entry R18 of the shared reference library. [S6] is not a link. It stays on the page as you see it, brackets and all, and names entry S6 in that product’s sources.md for you to look up.

That asymmetry is deliberate, and it is the same line the sources.md files draw between their own sections. A regulatory or actuarial reference is an authority the model is held to, and following it is part of reading the document. A primary product source is a specification citation — the Allgemeine Versicherungsbedingungen, Produktinformationsblatt or Basisinformationsblatt a number was taken from — which says where a figure came from rather than what the model must obey. So one reads as a tag on the page and the other as a link off it.

Numbering is per product — S1 is a different source in each — so tags resolve against the document’s own product rather than one global list.

Tag	On the page	Meaning
[S#]	bracketed text	Fact taken from a primary product document (Allgemeine Versicherungsbedingungen, Produktinformationsblatt, Basisinformationsblatt (PRIIP-KID), Verbraucherinformation, Tarifblatt, Musterrechnung) listed in the product’s sources.md
[R#]	link	Fact taken from a product-specific regulatory/actuarial reference in the product’s sources.md
[REG-R#]	link	Fact taken from the cross-product reference library (frozen R-numbering)

std — a standardization introduced for the reference implementation: a parameter or convention chosen where sources vary, are proprietary, or are silent. Each carries a rationale and, where available, the observed range across insurers.

unverified — a claim that no retrieved document confirms. Treat it as a to-verify item, not an established fact. The tag was applied at drafting to every specific paragraph number, effective date, monetary amount and market figure, because at that point nothing had been read; the re-verification then discharged it wherever a document was opened and supported the claim, so a tag that survives now means the document was consulted and did not settle the point — or could not be opened at all.

The hard rule throughout: every quantitative parameter is either source-tagged or marked std. In this library that rule does most of its work on the biometric bases, which are std proxies throughout because the DAV tables are proprietary, and on the charge and premium levels, because German pricing is quote-driven and no public rate card exists.

Regulatory and actuarial reference library

The reference library is the curated cross-product bibliography — frozen numbering R1–R56, cited as [REG-R#] — with a product-relevance matrix. It spans the prudential frame in two layers, the European one (Solvabilität II, the Delegierte Verordnung (EU) 2015/35, the 2025 review directive and the EIOPA risk-free term structures) and the Versicherungsaufsichtsgesetz it reaches German life business through (VAG 2016 and its Anlage 1, the §§ 74–110 balance sheet and SCR/MCR, the §§ 124–125 Anlagegrundsätze with the Sicherungsvermögen and the Anlagestock, § 138 on Prämienkalkulation, § 139 on the Überschussbeteiligung and the Sicherungsbedarf test, the RfB provisions of §§ 140 and 145, the Verantwortlicher Aktuar and the 1994 deregulation, Protektor and the supervisor’s crisis powers, and the Solvency II transitionals); the reserving regulation and its rates (DeckRV § 2 and the Höchstrechnungszins, the rate history and the Sechste Verordnung of 19 July 2024, § 4’s Höchstzillmersätze, and § 5 Abs. 3’s Referenzzins, Zinszusatzreserve and Korridormethode); the surplus regulations (the MindZV’s 90/90/50 minima, the RfBV, the LVRG 2014, and BaFin’s MaGo and Auslegungsentscheidungen); the contract law of the VVG (the statute and § 171’s halbzwingende Vorschriften, the two Widerrufsrechte, § 153 and the hälftige Beteiligung an den Bewertungsreserven, the Modellrechnung and Standmitteilung, the Selbsttötung and beneficiary sections, § 163’s Prämien- und Leistungsänderung, §§ 165–170 on the paid-up right, the Rückkaufswert and the Stornoabzug, and Kapitel 6’s §§ 172–177 on Berufsunfähigkeit); conduct and disclosure (the VVG-InfoV and Effektivkosten, PRIIPs, the IDD, Test-Achats with the AGG, and the 2023 Wohlverhaltensaufsicht Merkblatt); the BGH line of authority and the GDV Musterbedingungen; the tax architecture of the Drei-Schichten-Modell (the AltEinkG, the Basisrente deduction and its five prohibitions, the ZPO Pfändungsschutz that shapes them, the Ertragsanteil and Besteuerungsanteil, the Riester machinery of § 10a and §§ 79–99, the AltZertG with the BZSt and the Produktinformationsstelle, the 2026 reform and the Altersvorsorgedepot, the 12/62 rule and the Mindesttodesfallschutz, and the ErbStG with the SGB V contributions on an annuity in payment); the biometric bases and market statistics (Rechnungsgrundlagen erster und zweiter Ordnung and the DAV as owner of the tables, DAV 2008 T, DAV 2004 R and DAV 2004 R-Bestand, the DAV 1997 Berufsunfähigkeit family, DAV 2008 P with the § 15 SGB XI Pflegegrad break, Destatis, and the market-in-numbers entry); and the accounting and professional standards (HGB §§ 341–341o with the RechVersV and BerVersV, IFRS 17 and the Variable Fee Approach, and the DAV Fachgrundsätze with the annual Höchstrechnungszins recommendation).

The product-relevance matrix runs the fifty-six entries against the ten products in three states — load-bearing, qualified or background, and not relevant — so a reader can see at a glance that R51 (DAV 2008 P and the Pflegegrad break) is load-bearing for one product and background for three, while R34 (unisex) and R47 (the first- and second-order bases) are load-bearing for all ten. One instrument is deliberately absent from the matrix and carries no id: BaFin’s Kapitalanlagerundschreiben and the Anlageverordnung it interprets bind small insurers under §§ 212–217 VAG and domestic Pensionskassen and Pensionsfonds, not the Solvency II life insurers that write these ten products, which are governed by the qualitative § 124 VAG prudent person principle — German market writing routinely cites AnlV quotas as though they bound everyone, and the circular is discussed inside R7 so that no delib author misapplies one. Read the page’s own retrieval-conditions section first: it is stated in full at the head, before the first entry, and every one of the fifty-six entries records Fetched: no.

Known gaps and caveats

Aggregated from the per-product research; each product’s documents carry the full list, and each _research/<slug>.md closes with its own numbered register.

• The library was drafted blind, then re-verified; coverage of that re-verification is uneven. Direct HTTP egress was blocked while delib was written — gesetze-im-internet.de, bafin.de, aktuar.de, gdv.de, destatis.de, dejure.org, eur-lex.europa.eu and de.wikipedia.org each refused with HTTP 403 at the gateway — and the 200-call WebSearch budget was spent on a prudential sweep and a contract-law sweep, leaving the tax and biometric sweeps and eight of the ten product files with no research channel at all. That first draft rested on the authoring model’s own knowledge. The policy has since been lifted and the citations re-checked: 613 of 969 source entries now read Retrieved: yes, 37 were reached in part, and 319 could not be opened — 404 at the cited address, a consent or JavaScript wall, a paywall, a subscription login, or an address that was never established. The statutory core is near-complete; the thinnest products are those whose primary sources are carrier wordings behind a document portal, fondsgebundene_rentenversicherung most of all at 25 of 51. An entry that still reads Retrieved: no is a pointer, not a certificate.

• The re-verification corrected the drafted text, so the drafted text was wrong in places. Sourcing, figures and attributions moved in every product. The sharpest case: the klassische_rentenversicherung corpus had leaned on CosmosDirekt LA 904 A citing DAV 2004 R at 0 percent p.a., and the retrieved document contains neither — it states “der tarifliche Garantiesatz von 0,90 Prozent p. a.”, which is its 2017 vintage’s cap. Both facts are now sourced to documents that carry them. Elsewhere: DAV 2008 T observes 2001–2004, not 2006–2008; the GDV Stornoquote is 2,56 % for 2023; three kapitallebensversicherung sources cited as endowment wordings are annuity wordings. Each product’s sources.md records its own corrections. Findings that reach a modelled rate or rule were reported and not acted on — a model change moves the worked example and its golden tests — and each product’s model.md or sources.md names them.

• The DAV tables are proprietary, so every biometric basis here is a std proxy. DAV 2008 T, DAV 2004 R and DAV 2004 R-Bestand, DAV 1997 I / RI / TI and DAV 2008 P are the property of the Deutsche Aktuarvereinigung, are not public, and are cited by name and never redistributed REG-R47 REG-R48 REG-R49 REG-R50 REG-R51. Each shipped table is anchored so the model’s worked example reproduces exactly and each Data docstring says what a replacement must preserve, which is the honest form of the claim — but no delib decrement rate is a market observation. Two further problems are specific rather than general: no public German BU or Pflege decrement table exists in any form, so BU_DE_S’s inception, reactivation and disabled-mortality bases and Pflege_DE_S’s whole transition matrix are constructions whose shapes are argued and whose levels are invented; and DAV 2008 P was built on the superseded Pflegestufen, which the 2017 Pflegegrad reform replaced in a way that widened the insured population — the largest basis risk in that product, and one a licensed copy of the table would not by itself remove. Whether a successor to DAV 1997 I exists could not be established, and the DAV table names used in the BU file are themselves marked unverified.

• There is no public German rate card, and no price point of any kind, for any of the ten products. frlib had one published attained-age grid that TD_FR_S reproduces exactly; delib has none. Every premium in this library is computed by the model’s own equivalence principle on std bases and no delib premium reproduces a published figure. The single highest-value missing datum is named in the term-life register: one published Bruttobeitrag / Zahlbeitrag pair at a known age, sum insured and term would pin the Sicherheitszuschlag directly and re-derive the rest of the scale. The document class that would supply it — a Produktinformationsblatt — was not located for any product, in specimen or model form, and the one PRIIP Basisinformationsblatt anywhere in the corpus is for an endowment and its figures were not established.

• Charge levels are std throughout, and on one product they are structurally undisclosed rather than merely unretrieved. Not one Effektivkosten value, Abschlusskostenquote, Verwaltungskostenquote or commission rate was established for any carrier on any product; the only commission figure in the whole corpus is one carrier’s 25 ‰, and BaFin’s statement that Effektivkosten “differ considerably” and that it will examine outliers is qualitative. On term life the levels would have been missing even with full egress: there is no Effektivkostenquote because there is no yield, no Basisinformationsblatt because the product is not a PRIIP, and the Produktinformationsblatt quotes premiums rather than loadings. Every model therefore sits its acquisition charge exactly at the § 4 DeckRV Höchstzillmersatz of 25 ‰ of the Beitragssumme REG-R16, which demonstrates the binding constraint instead of inventing an interior point — and which may well be wrong for a slim direct-channel tariff, as the term-life register says in terms.

• No Rentenfaktor level, range or time series exists anywhere in the corpus — for any product, at any carrier, in any year. The rating house’s own article titled with the question returned no level, and the Rentenfaktor-Check titled as data and analysis returned none. So every factor in the library is a construction with its derivation printed beside it: 32,00 € at age 67 on RV_DE_S, 25,00 € on FRV_DE_S derived as 10 000 / (12 · T_eff) at a 0 % Rechnungszins rather than observed, 31,50 € on Basis_DE_S, 29,00 € guaranteed on Riester_DE_S, 25,00 € on Index_DE_S. Both branches of max(garantierter, aktueller) are exercised by a shipped model point on RV_DE_S and Basis_DE_S — the anchor cell converts at the current factor on each and model point 13 at the guarantee — and that is the point of choosing those two levels. On the other three the base run does not separate the branches: Riester_DE_S’s guaranteed 29,00 € wins on all thirteen points, and the current factor is lifted above it only by a test that substitutes a doubled annuitant mortality table; FRV_DE_S ships the two equal on twelve points and the current factor above the guarantee on point 13, so its guarantee never strictly binds; and Index_DE_S’s two factors are Projection References both fixed at 25,00 € rather than model-point columns, so neither branch is distinguished anywhere in the shipped run. None may be quoted as a market rate. Index_DE_S and RV_DE_S additionally record that their std Rentenfaktor and their std annuity table are not calibrated to each other, which is why the annuity there is reported and not computed.

• No behavioural rate is sourced, on any product. No German Stornoquote was established for any of the ten at any duration; the two GDV market-wide measures for 2024 — 2,72 % and 1,2 % — are irreconcilable from the search evidence, neither is product-specific or by duration, and RLV_DE_S deliberately declines to use either. Every lapse table ships with its shape argued from structure and its levels invented: the § 20 Abs. 1 Nr. 6 EStG twelve-year threshold puts a visible step at duration 12 on four models, FRV_DE_S keys the same step on age as well as duration because keying it on duration alone fires fourteen years early, and BU_DE_S’s unusually low level is a product fact rather than optimism, cover being unreplaceable once health has changed. Nor is any take-up rate established: no Beitragsfreistellung rate, no Zuzahlung utilisation, no Kapitalwahlrecht or Teilkapitalauszahlung take-up, no Nachversicherungsgarantie exercise pattern, and no election distribution for Index_DE_S’s annual Wahlrecht in either direction. Where a rate was needed the mechanic is carried as a deterministic model-point election instead, and each model says so.

• Two products rest on no carrier document at all. For the Indexpolice, no Bedingungswerk for any index tariff was obtained — the file’s own central defect — so the Indexjahr definition, the observation dates, the payoff wording, the base of the participation, the Wahlrecht timing and any Mindest-Cap are all written from knowledge of the design family; no cap level for any insurer in any year was established, and no documented worked Indexjahr was found, so the two the model reproduces are constructed, labelled std in every cell and wired into the shipped return path precisely so the mechanic is asserted rather than described. For the Pflegerentenversicherung, no PIB, IPID or BIB, no premium, no charge level, no lapse rate, no rating-agency wording analysis and no count of contracts in force was obtained — the GDV life series does not carve the product out and the PKV series excludes it by construction, it being life business.

• Every statutory paragraph number in this library is unverified. gesetze-im-internet.de was refused at the gateway, so no Fassung line was ever seen: every statutory statement here is current in substance as reported in August 2026 and none is version-pinned. The specifically open items are named rather than smoothed over — whether § 169 Abs. 1 VVG’s certain-event limitation is the route by which a term assurance has no surrender value; the internal paragraph structure of § 169 for the Zeitwert branch, so that no delib document cites a subsection for it; § 168 Abs. 3 VVG’s ending of the termination right at Rentenbeginn, on which the whole Sofort_DE_S specification rests; the MindZV section numbers, where the three percentages are used and no section number is cited anywhere; the exact range of sections § 176 VVG imports, on which the BU surrender value, paid-up right and Überschussbeteiligung all depend; and whether a pure-risk Pflegerente falls inside § 169 at all. Four VVG provisions the endowment depends on — §§ 37, 38, 150 and 152 — were never searched at all, and nothing is asserted about any of them anywhere in the library.

• Two source errors are recorded rather than quietly applied, and one document is excluded. A search summary in the corpus conflates § 169 Abs. 3 VVG’s five-year spreading of acquisition cost with the DeckRV’s 2,5 % Höchstzillmersatz: they are different rules with different functions — a floor on the value against a cap on the charge — and KLV_DE_S asserts them in two separate checks for that reason. The sofortrente research file puts the nachschüssig effect at about 5 %; it is 0,34 % on a monthly annuity, the 5 % being an annual-annuity identity applied to a monthly one — the research file is frozen and never amended, so the correction lives in that product’s technical notes. And an Austrian ERGO AVB returned by a German AVB search is excluded and recorded: the VVG, the DeckRV and the MindZV do not apply to it, and the same caution applies to any .at or .ch document a later search returns.

• Market figures are qualitative throughout, and one legal position could not be determined. No endowment-specific new-business or in-force series and no time series showing the effect of the 2005 tax change; no size for the index-participation segment, which GDV counts inside conventional annuity business and for which probably no separate figure exists; no Riester contract count, chassis split or new-business series, and no official statistic for ruhende Verträge at all; no Sofortrente volume, average Einmalbeitrag or purchase-age distribution; no LTC-annuity contract count. Nineteen of the twenty-six carriers named in the endowment brief produced no document of any kind, and the variations tables across the library are records of absence rather than of variation. The status of the private Altersvorsorge reform as at the access date could not be established beyond a 2024 draft creating an Altersvorsorgedepot that did not become law in that parliamentary term, so nothing in this library asserts the current legal position on it.

• unverified items remain wherever a claim could not be corroborated by any search result, and under these retrieval conditions the tag does more work than it does in the sister libraries. It sits on the intermediate Höchstrechnungszins history between 1994 and 2025 and on every effective date in it; on the DeckRV amendment’s Bundesgesetzblatt year, inferred from the surrounding chronology and carrying no BGBl citation, none having been returned and none being invented; on every Ertragsanteil value but the 18 % at age 65; on the § 851c ZPO protected amounts, which the Basisrente file states the shape of and prints not one level of; on the Landgericht Köln Rentenfaktor decision, which is reported with no case number, date, parties or appeal history, so no delib document gives it a docket; and on the 2025 statutory Pflege benefit amounts, whose 2026 position is unknown.