Admissible Obligation Transitions
Abstract
An unpaid obligation must remain identifiable when its holder, terms, or software representation changes. A payment must discharge the right obligation once, and a later return must preserve its history. We define a shared procedure for changing obligations. Each kind of obligation supplies its state and update rules. An execution engine orders competing updates, checks authority and evidence, and records each decision. Under explicit assumptions about those rules and their execution, accepted updates preserve the stated safety conditions. Each recorded payment or other external event can change obligations at most once within one claim. Retained inputs reproduce past decisions, including refusals and requests for missing evidence. Shared records keep the total credited across claims within the supported payment amount. Retained obligation histories preserve rights, payment attribution, decisions, and unfinished instructions when records split, combine, or move to another program. Under the stated structural conditions, positive rational position partitions also preserve unfinished computations across later observations. The original ledger continues to govern payments during that computation. For obligations already resolved to payable amounts, consecutive quantity portions can be represented together while preserving their order. This gives the same payment and return behavior without listing every unit at the chosen monetary precision. The results depend on the stated authority, evidence, and storage assumptions. Applicable law determines the legal effect of the recorded changes.
The author has a commercial interest in systems of the kind this paper describes.
1 What must survive a change to a claim
An issuer owes a holder 100 units. A final payment of 60 leaves 40 unpaid. Suppose the record is then split into a paid position of 60 and an unpaid position of 40. The issuer still owes 40, and the earlier payment must remain attached to the portion it discharged. A transfer can change who may collect that remainder. An authorized amendment can change its due date or other terms. Moving the record into another program must retain each of those acts and the unpaid obligation that results.
The same history must distinguish a payment from its later return. If returned funds restore part of the obligation under the governing terms, a replacement payment has its own evidence. The engine must retain the first payment, the return, and the replacement. A current balance alone cannot show which payment a later correction concerns.
A financial claim records the right to performance. Its corresponding obligation records what a debtor must do. Besides quantity, these records need parties, conditions, dates, transfer restrictions, amendment authority, and recovery rules. We call claims governed by one such state model and rule set a claim family. Different families can share the operations of issuance, transfer, evidence admission, and settlement while giving those operations different effects.
A separate event engine for every family duplicates authority checks, settlement records, and audit logic. A common transferable balance retains quantity but omits such matters as coupon dates, insurance conditions, and receivable defenses. The question is which rules a common engine must enforce while each family determines its own obligations.
We define an admissible obligation transition as a family derivation accepted by the engine under both sets of laws. A derivation records a proposed change and the premises that justify it. The engine checks whether those premises permit the change from the current record. It also retains the reason when a request fails or needs external evidence.
The first result concerns one claim under fixed semantics. Later constructions retain payment attribution across claims and the identity of duties across changes of representation. The distinction between amendment and representation change remains explicit: the former changes rights under authority, while the latter preserves their recorded meaning.
Contract combinators already give precise and compositional payoff descriptions [28]. ACTUS specifies contract types through state variables, scheduled events, and cash-flow functions [1]. The Common Domain Model gives standardized product and lifecycle representations [10]. This paper addresses a different boundary. It asks which conditions let one state-transition engine operate across such families without erasing authority, unsettled obligations, holder rules, or external finality.
2 One obligation from issue to payment
A zero-coupon obligation has one payment at maturity and no intermediate coupon payments. Issuer I promises to pay initial holder A 100 units on day 30. The claim begins in draft state. The following eight events distinguish changes in the obligation from changes in the evidence about it.
- (X1)
-
I issues the claim to A under a valid signing authority.
- (X2)
-
A transfers the claim to eligible holder H.
- (X3)
-
H requests a transfer to Y, whose eligibility evidence is missing. The evaluator returns \mathsf{Awaiting} and names the required authority.
- (X4)
-
H attempts to transfer the claim to ineligible holder X. The evaluator refuses the request.
- (X5)
-
On day 30, a payment provider gives evidence p_{60} that 60 units have settled finally.
- (X6)
-
A request attempts to use p_{60} a second time. The evaluator refuses the request.
- (X7)
-
The provider reports that another 40 units are accepted but still pending.
- (X8)
-
The provider later gives evidence p_{40} for final settlement of the last 40 units.
To record these events, keep three pieces of history beside the current issuer, holder, and face amount. The map M records amounts used for final settlement. The ledger L records each payment’s evidence status, including pending observations. The set C records which payment occurrences have already been used. A key k identifies the payment occurrence described by provider evidence p; two messages about the same payment retain the same key. Subscripts in the trace name the quantities recorded so far.
The state trace is:
\begin{aligned} \mathsf{Draft} &\longrightarrow \mathsf{Outstanding}(I,A,100;M_0,L_0,C_0) \\ &\longrightarrow \mathsf{Outstanding}(I,H,100;M_0,L_0,C_0) \\ &\xrightarrow{\text{awaiting }Y} \mathsf{Outstanding}(I,H,100;M_0,L_0,C_0) \\ &\xrightarrow{\text{refused }X} \mathsf{Outstanding}(I,H,100;M_0,L_0,C_0) \\ &\xrightarrow{p_{60}} \mathsf{Outstanding}(I,H,100;M_{60},L_{60},C_{60}) \\ &\xrightarrow{\text{refused reuse}} \mathsf{Outstanding}(I,H,100;M_{60},L_{60},C_{60}) \\ &\xrightarrow{\text{pending }p_{40}} \mathsf{Outstanding}(I,H,100;M_{60},L_{60,40},C_{60}) \\ &\xrightarrow{\text{final }p_{40}} \mathsf{Discharged}(I,H,100;M_{100},L_{100},C_{100}), \end{aligned} where M_0=L_0=C_0=\varnothing, M_{60}=\{k_{60}\mapsto60\}, L_{60}=\{k_{60}\mapsto\mathsf{Final}\}, L_{60,40}=L_{60}\cup\{k_{40}\mapsto\mathsf{ProviderAccepted}\}, and M_{100}, L_{100}, and C_{100} contain both settlement occurrences. The outstanding amount is the face amount minus the sum recorded in M.
The private state is unchanged by the awaiting and refusal results. The provider-accepted observation changes only the settlement-evidence ledger. It does not change the obligation. Final-settlement evidence can contribute to an obligation change only once. A provider acceptance is not final settlement.
Issuance needs authority, and transfer needs evidence that the proposed holder may receive the claim. Partial settlement reduces the outstanding amount by exactly the final amount. Keeping evidence status separate lets the engine record a pending payment while the obligation remains due. Retaining each decision lets an auditor reconstruct both changes and refusals.
A common evaluator needs the family to expose these distinctions without requiring the same private state for every product. The interface below makes that requirement precise. The formal zero-coupon rules later instantiate its transition-critical part.
3 The family and engine interfaces
The family determines which rights an event changes. The engine determines which request and evidence it evaluates, and which decision becomes part of the history. We first specify external evidence because both depend on it.
A commit records one decision and its writes indivisibly. Its linearization point is the instant at which the model orders that commit relative to competing operations. Ingress is the preliminary check that a request may enter this decision process. A base identifies the current record against which the request was prepared. A nonce identifies one submission so that a retry can recover its existing result.
A content digest identifies immutable encoded content. Throughout the paper, \mathsf{Enc} denotes canonical encoding, H its commitment hash, and d(v) the content digest of a value v. A bare d later denotes an economic duty. These uses are distinguished by the argument of the digest notation. The notation \mathcal{P}_{\mathrm{fin}}(X) denotes finite subsets of X, \uplus denotes a disjoint union, and \rightharpoonup denotes a partial map.
3.1 External facts and commit-time status
Some facts come from outside the claim engine. A public register may establish who holds an office. A separate instrument, rule, and current status must establish whether that office carries authority for the requested transaction. That authority can end by law even if the office record remains unchanged. A compliance service may establish holder eligibility. A custodian may establish asset control. A payment system may establish final settlement. A tribunal or court may establish a disputed outcome.
Fix ambient types \mathsf{Principal}, \mathsf{Time}, and \mathsf{Snapshot}. Fix also a global type \mathsf{OccurrenceKey} of issuer-authenticated identifiers for external occurrences. A family admits such an identifier only through its own typed canonicalization map.
The payment example needs an identity for the credit itself as well as an identity for each assertion about it. An amount correction changes an assertion about the credit. It must not create another credit that the engine can use for a second discharge. The following record separates those identities.
Definition 3.1 (External fact).
An external fact is a tuple \begin{aligned} e=(&\mathit{factId},\mathit{issuer},\mathit{subject},\mathit{predicate}, \mathit{scope},\mathit{eventTime},\\ &\mathit{validity},\mathit{effectKey},\mathit{evidence}). \end{aligned} The optional \mathit{effectKey}\in\mathsf{OccurrenceKey} is an issuer-authenticated identifier for one external occurrence in one admitted authoritative namespace. For a payment, the admission policy nominates the actual recipient-credit record: k=(\mathit{authoritativeNamespace},\mathit{recordKind}, \mathit{stableEventId}). Provider messages and reporting wrappers resolve to this record through authenticated aliases. The policy checks its book, account, asset, and identifier epoch. A provider namespace alone does not establish a distinct credit. Distinct postings remain distinct even when they share a command. The claim, amount, and currency are authenticated attributes of the occurrence, not parts of its identity. The legal origin, reporter, observation time, and command identifier also remain separate. Different assertions about one occurrence carry the same effect key even when an attribute is corrected. The assertion identifier distinguishes messages. The effect key distinguishes the occurrence those messages describe.
Write \mathsf{Fact} for the type of well-typed tuples in this definition.
The definition does not make an external fact true. It makes the asserted source, scope, occurrence, and evidence explicit. The issuer remains responsible for the identifier namespace that it authenticates.
An effect premise is an admitted fact asserting an external occurrence that an obligation transformer consumes. A settlement-observation premise reports the evidentiary state of a settlement occurrence without changing the obligation. A status premise asserts a condition such as authority, capacity, or eligibility. Every single-use external effect must be expressed as a keyed occurrence. An authority fact used to issue a claim is a status premise. A provider assertion below the route’s final class is a settlement-observation premise. A payment used to discharge the claim is an effect premise.
Authority can change while a request is being evaluated. The engine therefore needs an authenticated view of the relevant sources at commitment. A source head identifies its current recorded state; the snapshot records the heads used by this decision.
Definition 3.2 (Authenticated status snapshot).
At commit time \tau, a status service publishes an authenticated snapshot \begin{aligned} \Sigma_\tau=(&\mathit{heads},\mathit{issuerStatus},\mathit{credentialStatus}, \mathit{capacityStatus},\\ &\mathit{insolvencyStatus},\mathit{revocations},\mathit{policyStatus}, \mathit{observedAt},\mathit{headAttestation}). \end{aligned} The map \mathit{heads} names the source head for each predicate. Capacity and insolvency status can come from public authorities other than the issuer. The head attestation authenticates the listed source heads as current at the commit linearization point. Its content digest h_\Sigma binds the status information and attestation used there. The admission policy maps each predicate to one named source, a maximum snapshot age, and a rule for current non-revocation.
Definition 3.3 (Admitted fact set).
Fix a content-addressed admission policy h_\Pi, claim x, event a, asserted event time \theta, sequencer-assigned commit time \tau, and snapshot \Sigma_\tau. The finite set E_{\Pi,x,a,\theta,\tau,\Sigma} contains exactly the proposed facts that pass source, authentication, subject, scope, event-time validity, commit-time status, and revocation checks. The policy explicitly states whether delayed recording is permitted. An actor-supplied timestamp cannot create that permission. After the engine obtains an authenticated current snapshot, a missing party-supplied premise can return \mathsf{Awaiting} with the named requirement. Failure to obtain or authenticate a mandatory current source is an ingress failure, not \mathsf{Awaiting}.
Every admitted event time satisfies \theta\leq\tau. A policy may bound how far into the past an event may be recorded; it cannot admit a future event.
The admitted set has a canonical order and digest. Admission proves only that the named policy allowed reliance on the assertion for this request under h_\Sigma. It does not prove the assertion or the policy’s legal adequacy.
3.2 Claim families
The interface groups the zero-coupon example’s requirements into explicit mathematical data. State, events, and the safety predicate describe valid local behavior. The obligation and holder projections expose what a transition changes. A derivation then separates effects justified by external occurrences from changes, such as issuance, that its authority permits directly.
The remaining data record consumption, settlement evidence, and permitted reliance on receipts. Settlement evidence uses a preorder because two different assertions can have equal evidentiary strength. Its least element means that no evidence has been admitted. A finite map records the exceptional entries; all other entries retain that least value.
Definition 3.4 (Claim family).
A claim family F consists of:
- (F1)
-
a family identifier and a content digest h_F of its canonical semantics;
- (F2)
-
a state space \mathcal{S}_F with initial states \mathcal{S}_F^0 and an event alphabet \mathcal{A}_F;
- (F3)
-
a trace-aware derived-event schedule \delta_F(T_{<i},s,\tau)\in\mathcal{P}_{\mathrm{fin}}(\mathsf{DerivedId}_F\times\mathcal{A}_F) and a well-founded priority order for its pending events, where the first projection of the returned set is injective;
- (F4)
-
a safety predicate P_F:\mathcal{S}_F\to\{0,1\};
- (F5)
-
an obligation projection B_F:\mathcal{S}_F\to\mathcal{O}_F;
- (F6)
-
a holder projection G_F:\mathcal{S}_F\to\mathcal{P}_{\mathrm{fin}}(\mathsf{Principal});
- (F7)
-
a derivation type \Pi_F and proof-relevant transition relation U_F\subseteq \mathcal{S}_F\times\mathcal{A}_F\times\mathcal{P}_{\mathrm{fin}}(\mathsf{Fact})\times\mathsf{Time}^2 \times\mathsf{Snapshot}\times\mathcal{S}_F\times\Pi_F;
- (F8)
-
types \mathsf{EffectUse}_F, \mathsf{EffectKey}_F, and \mathsf{EffectDelta}_F; a partial canonicalization \kappa_F:\mathsf{OccurrenceKey}\rightharpoonup\mathsf{EffectKey}_F; maps \begin{gathered} \mathsf{key}_F:\mathsf{EffectUse}_F\to\mathsf{EffectKey}_F, \qquad V_F:\Pi_F\to\mathcal{P}_{\mathrm{fin}}(\mathsf{EffectUse}_F),\\ \Delta_F^{\mathrm{eff}}:\Pi_F\to (\mathsf{EffectKey}_F\rightharpoonup\mathsf{EffectDelta}_F), \qquad C_F:\mathcal{S}_F\to\mathcal{P}_{\mathrm{fin}}(\mathsf{EffectKey}_F), \end{gathered} where each partial map \Delta_F^{\mathrm{eff}}(\pi) has finite domain;
- (F9)
-
a canonical total order on effect keys, an effect-independent transformer D_F^0:\Pi_F\to(\mathcal{O}_F\to\mathcal{O}_F), an effect application R_F:\mathsf{EffectDelta}_F\to(\mathcal{O}_F\to\mathcal{O}_F), and the obligation transformer D_F defined by D_F(\pi)=R_F(\Delta_m)\circ\cdots\circ R_F(\Delta_1)\circ D_F^0(\pi), where k_1<\cdots<k_m is the ordered domain of \Delta_F^{\mathrm{eff}}(\pi) and \Delta_j=\Delta_F^{\mathrm{eff}}(\pi)(k_j);
- (F10)
-
an authority predicate A_F:\Pi_F\to\{0,1\} and holder predicate H_F:\Pi_F\times\mathsf{Principal}\to\{0,1\};
- (F11)
-
a distinguished settlement-key subtype \mathsf{SettlementKey}_F\subseteq\mathsf{EffectKey}_F, and, for each k\in\mathsf{SettlementKey}_F, a pointed evidence preorder for its route, (\mathcal{E}_{F,k},\preceq_{F,k},\mathsf{None}_{F,k}) with a route-defined element \mathsf{Final}_{F,k}, together with the total maps L_F:\mathcal{S}_F\to\prod_{k\in\mathsf{SettlementKey}_F}\mathcal{E}_{F,k}, \qquad Q_F:\Pi_F\to\prod_{k\in\mathsf{SettlementKey}_F}\mathcal{E}_{F,k}, each with finite non-bottom support;
- (F12)
-
a reliance-scope preorder (\mathcal{Y}_F,\sqsubseteq_F);
- (F13)
-
a finite enumeration \mathsf{ReasonTag}_F and a typed payload schema for each refusal tag; and
- (F14)
- a finite enumeration \mathsf{RequirementTag}_F and a typed payload schema for each missing-evidence tag.
The family lists the individual external uses in a derivation. Their image under the key map is the set of occurrences consumed by that step:
Write W_F(\pi):=\mathsf{key}_F[V_F(\pi)]. The obligation carrier depends on the family. It can be one amount, a dated multiset, or a set of covered duties. A transition derivation \pi exposes the authority, holder, and external-effect premises used by that transition. The common engine verifies the derivation but does not interpret the private obligation carrier.
The settlement ledger is indexed only by settlement effects. Tribunal, corporate-action, insurance, and other obligation-changing occurrences remain general effect keys; they need not have settlement-ledger entries. The ledger is not one scalar status for the whole claim. A final 60-unit payment and a pending 40-unit payment therefore occupy different keys. For a settlement key k on one route, the evidentiary order can contain \mathsf{None}_{F,k}\prec_{F,k}\mathsf{Submitted} \prec_{F,k}\mathsf{ProviderAccepted}\prec_{F,k}\mathsf{Final}_{F,k}. \mathsf{Final}_{F,k} means that the route’s content-addressed status map classifies the settlement as irrevocable and unconditional under that route’s rules and legal basis. A reversible provider success cannot map to \mathsf{Final}_{F,k}. The preorder compares evidentiary strength. It does not replace the route’s temporal state machine.
A reliance scope names the class of relying party, the proposition that the receipt supports, the jurisdiction or rule context, and an expiry. The order \sqsubseteq_F means “no broader than.” A technical receipt can therefore support engine audit without purporting to create contractual, regulatory, or third-party legal reliance.
One family can govern many individual claims. Each instance binds the exact semantics that currently interpret its state. A template pack, defined below, adds the institutional conditions for using the family in a named setting.
Definition 3.5 (Claim instance).
Let \beta=(h_F,h_T,h_R,h_\Pi) be an active semantic binding. A claim instance is x=(\mathit{id},F,\beta,s_0), where \mathit{id} is unique and s_0\in\mathcal{S}_F^0. The four digests resolve to immutable family, template-pack, rule, and admission-policy content in the engine registry.
Some events become due without a fresh human instruction, such as a coupon payment date or milestone review. The scheduler must recognize the same due event after a retry or a software change. Otherwise it could create another execution opportunity for an already decided event.
A scheduled occurrence has an enduring economic duty identifier d, admitted from its constituting act and occurrence index. Section 7 defines its canonical registry and migration rule. For event class c and authorized decision generation g, define u=H(\mathsf{Enc}(d,c,g)). The active semantic binding separately identifies the program that evaluates this event. A family due key resolves to its registered d, never to a new duty merely because a content digest changes. The trace gives each derived identifier a disposition in \{\mathsf{Pending},\mathsf{Accepted},\mathsf{Refused}\}. Every due pair returned by \delta_F uses the displayed identifier. A due identifier with no terminal decision is pending. The schedule returns no terminal identifier. Awaiting leaves the identifier pending. Acceptance and refusal are terminal for that identifier. Reactivation after refusal requires an authorized new decision generation for the same duty. It preserves the refusal and the economic history.
A derived event does not mutate a claim inside another event’s atomic step. The authorized scheduler submits it through the ordinary transition interface with origin \mathsf{Derived}(u). Each derived event therefore has its own request, decision, result, receipt, and trace step. The engine recomputes \delta_F after every committed step and orders only pending due identifiers. After an awaiting result, a retry uses the same derived identifier, a fresh nonce, and the current base. A conflicting authored event cannot pass a pending higher-priority derived event. Evidence input and cure remain available while that event awaits. Section 3.6 gives their transitions and progress conditions.
3.3 Family admission packs
A family definition states possible claim behaviour. It does not state that a particular institution may issue, transfer, settle, or rely on that claim in a particular legal setting. That second question belongs to an admission pack.
Definition 3.6 (Template pack).
A template pack for a family F is the tuple \begin{aligned} T_F=(&h_F,h_R,h_\Pi,\mathsf{Legal}_F,\mathsf{Auth}_F, \mathsf{Recognition}_F,\\ &\mathsf{Evidence}_F,\mathsf{Risk}_F,\mathsf{SettlementAsset}_F, \mathsf{NettingBasis}_F,\\ &\mathsf{PhaseGate}_F,\mathsf{Dispute}_F, \mathsf{Report}_F,\mathsf{Reliance}_F),\\ h_T={}&H(\mathsf{Enc}(T_F)). \end{aligned} The pack stores legal classification by named regime and purpose in \mathsf{Legal}_F. It stores issuance and lifecycle authorities in \mathsf{Auth}_F. For each regime and purpose, its recognition field \mathsf{Recognition}_F records the named authority or published rule, effective period, and one status in \{\mathsf{Recognized},\mathsf{Rejected},\mathsf{Unknown}\}. \mathsf{Evidence}_F names evidence sources, freshness rules, and retention duties. \mathsf{Risk}_F names risk limits, settlement routes, and their finality maps. \mathsf{SettlementAsset}_F supplies the eligibility record defined below. \mathsf{NettingBasis}_F records no netting, central-counterparty interposition by novation or open offer, or multilateral agreement without interposition; it also names the population, window, and purpose to which that basis applies. \mathsf{PhaseGate}_F is a Boolean predicate, signed by its named pack authority, over event phase and any holding period. \mathsf{Dispute}_F names dispute and recovery rules. \mathsf{Report}_F names reporting duties. \mathsf{Reliance}_F supplies the pack-specific permitted-scope function Z_{F,T}. The pack digest binds the complete tuple. Its internal digests bind the family, rule, and admission semantics used by the transition request.
\mathsf{SettlementAsset}_F is a finite set of records (\mathit{asset},h_P,\mathit{evidence},\mathit{reviewCycle}). The digest h_P binds a settlement policy. Its evidence must meet the policy’s current-use conditions.
For the pack’s admission check, an instrument’s harbor set lists jurisdictions recorded as admitting its constitution or operation. The issuer’s harbor set supplies the corresponding entity-level boundary. The following inclusion requires each instrument jurisdiction to lie within that boundary.
For an instrument S issued by an entity E, the pack also requires \mathcal H(S)\subseteq\mathcal H(E), where \mathcal H gives the set of jurisdictions in which the instrument or entity is harbored. A composite instrument needs an explicit recognition record for the composition. Recognition of each component does not imply recognition of the composite.
The Claim as Primitive defines the payoff program and pricing interface that each family instantiates [21]. One Entity in Many Jurisdictions defines the composed standing and instrument harbor set that constrain holder eligibility [22].
An obligation of 100 units needs a rule identifying which asset and transfer count as those 100 units of performance. That rule also determines the evidence needed for the intended amount and recipient. The settlement policy records these conditions explicitly, including conversion when the units differ.
Definition 3.7 (Settlement policy and eligibility).
A settlement policy P fixes the asset, unit, precision, route, authoritative book, regime, purpose, participant population, size limits, and finality map. Its asset basis has exactly two constructors: \begin{aligned} &\mathsf{RedeemableClaim}(\mathit{obligor},\mathit{redemptionRule},\mathit{clock}),\\ &\mathsf{NativeAsset}(\mathit{network},\mathit{assetRule},\mathit{controlRule}). \end{aligned} A redeemable claim requires its redemption obligation and enforceability evidence. A native asset requires its own control and transfer rules. It carries a typed absence of a redemption obligation. Missing redemption evidence for the first constructor cannot select the second.
Eligibility requires evidence for eight policy-indexed conditions:
- (E1)
-
the exact asset, unit, precision, route, purpose, and quantity;
- (E2)
-
the chosen asset basis, including obligor capacity and redemption rights when the basis creates those obligations;
- (E3)
-
permitted participants, recipients, custody arrangements, and their authority;
- (E4)
-
valuation, liquidity, and funding capacity for the intended size;
- (E5)
-
any conversion’s two units, quote, quantity, rounding rule, route, and expiry;
- (E6)
-
the authoritative booking and status map that supports discharge;
- (E7)
-
the funding, execution, and applicable redemption clocks, each at its named use stage; and
- (E8)
-
stress funding, contingency, substitution, and concentration rules.
A payment in the claim’s own native unit requires no fiat conversion. A payment in another unit requires the family’s evidenced conversion rule. Equal numerals in different units do not establish equal performance. An inapplicable condition has its typed justification in P. An applicable condition without evidence remains unresolved. Eligibility concerns the named use under h_P, rather than an asset name alone.
Netting can replace several gross obligations with a smaller set under a specified legal arrangement. Its arithmetic and its legal basis are separate requirements. The pack records the latter before any joint transformer can use it.
Definition 3.8 (Pack netting basis).
\mathsf{NettingBasis}_F has exactly three constructors: \begin{aligned} &\mathsf{None},\\ &\mathsf{CCPInterposition}(\mathit{mode},\mathit{population}, \mathit{window},\mathit{purpose}),\\ &\mathsf{MultilateralAgreement}(\mathit{population},\mathit{window}, \mathit{purpose}). \end{aligned} where \mathit{mode} is novation or open offer [29]. This field records the asserted legal basis. It does not prove enforceability or the cross-claim extension in Section 5.1.
Registration and legal recognition are different relations. A pack is registered when its fields are complete, mutually consistent, and accepted by the engine registry. It is recognized by a regime for a purpose only when a named authority or published rule gives the pack that effect. Recognition is indexed by both regime and purpose. \mathsf{Unknown} fails a recognition condition; it is neither recognition nor rejection. Registration alone creates no legal recognition.
Definition 3.9 (Pack admission).
An instrument governed by T_F is eligible for a requested transition only if all eight conditions hold at the linearization point:
- (A1)
-
the registry resolves h_T to one registered immutable pack and resolves every digest inside it to one immutable object;
- (A2)
-
the instrument and issuer fall within the pack’s legal profile for the named regime and purpose; when that transition requires recognition, a current record in \mathsf{Recognition}_F names the authority or published rule for that regime and purpose with status \mathsf{Recognized}, and any composite has its own recognized mapping;
- (A3)
-
each actor has the authority required for this transition;
- (A4)
-
every added or changed holder satisfies the applicable holder rule;
- (A5)
-
every evidence premise passes its source, scope, freshness, and non-revocation rule;
- (A6)
-
the requested event is an admitted edge from the current lifecycle state, the signed phase and holding-period gate is true, and every obligation, cap, harbor, and risk condition is satisfied;
- (A7)
-
each settlement effect uses an admitted route and eligible settlement asset and can reach no finality class above that route’s authenticated evidence; and
- (A8)
-
the derivation satisfies every reporting and dispute condition that the pack makes applicable to this transition.
Let \Omega_{F,T} be the type of pack-admission witnesses. The predicate \mathsf{PackAdmit}_F(T_F,s,a,E,\theta,\tau,\Sigma,\pi,\omega)=1 means that \omega\in\Omega_{F,T} establishes conditions (A1)–(A8) for the derivation \pi.
A guard is one condition that the request must satisfy. Ordered guards make a negative decision reproducible when several conditions fail or need evidence. The decision certificate records the first such condition and the facts used to classify it.
The content-addressed family and pack semantics declare a finite guard sequence. It covers family derivability and all applicable conditions (A1)–(A8). The order respects each guard’s typed input dependencies and fixes failure precedence. A canonical negative certificate \nu\in\mathsf{NegativeCert}_{F,T} names the first failed or missing condition and binds the facts that establish that classification. Define the total pre-receipt decision type \begin{aligned} \mathsf{Decision}_{F,T}={}&\mathsf{Accept}(\mathcal{S}_F\times\Pi_F\times\Omega_{F,T})\\ &\uplus\mathsf{Refuse}(\mathsf{Reason}_F\times\mathsf{NegativeCert}_{F,T})\\ &\uplus\mathsf{Await}(\mathsf{Requirement}_F\times\mathsf{NegativeCert}_{F,T}). \end{aligned} For fixed inputs, \mathsf{PackDecision}_F(T_F,s,a,E,\theta,\tau,\Sigma) \in\mathsf{Decision}_{F,T} returns the accept constructor when the family derivation and all eight conditions hold. It returns the refusal constructor for the first resolved false condition and the awaiting constructor for the first missing party-supplied premise after successful ingress. An unavailable or unauthenticated mandatory commit-current source is an ingress failure. The certificate of an accepted decision is \omega; otherwise it is \nu; this defines the projection \mathit{cert}(\eta). The permitted-scope function has type Z_{F,T}:\mathcal{S}_F\times\mathcal{A}_F\times\mathcal{P}_{\mathrm{fin}}(\mathsf{Fact})\times \mathsf{Decision}_{F,T}\to\mathcal Y_F. Z_{F,T} is evaluated after the decision exists and before its receipt is constructed.
The following catalogue shows how the pack question varies with claim meaning:
| Family | First-tier pack question |
|---|---|
| Equity | Which law, register, holder rule, and corporate-action authority bind? |
| Fixed income | Which payment schedule, amendment authority, and settlement route bind? |
| Structured and derivative | Which payoff, valuation source, collateral rule, and close-out basis bind? |
| Sharia-compliant | Which asset, undertaking, approval authority, and cash-flow form bind? |
| Real-world asset | Which title book, custodian, encumbrance rule, and delivery right bind? |
| Insurance and parametric | Which coverage, trigger, assessment, limit, and recovery rule bind? |
| Event-contingent | Which event source, resolution rule, and settlement obligation bind? |
| Loyalty and engagement | Which issuer promise, holder class, expiry, and redemption rule bind? |
| Intellectual property and royalty | Which right, territory, license, usage record, and royalty waterfall bind? |
Each entry calls for a family and one or more packs. The table describes possible admission questions; it does not supply their answers or prove the family laws. Intellectual-property and royalty claims use the final entry unless the instrument has another family’s legal and economic form.
A compliance tier groups states by a threshold in an order of compliance conditions. A state passes when it meets that threshold. Nested thresholds retain this ordering when conditions compose; an unexplained color label does not specify such a rule [20].
3.4 Serialized engine environment
Two valid requests may both name the same unpaid 40 units. The store must choose a serial order and let the second request see the first decision. Comparing the requested base with the current base supplies that check. A new revision distinguishes a later record even if its private balance returns to a value seen before.
Definition 3.10 (Engine environment).
An engine environment is \mathcal{G}=(\mathsf{Enc},H,\mathsf{Clock},\mathsf{Registry}, \mathsf{Status},\mathsf{EvidenceStore},\mathsf{Store},\mathsf{Eval}, \mathsf{Receipt}). \mathsf{Enc} is a canonical typed encoding and H is its commitment hash. \mathsf{Registry} maps content digests to immutable family, template-pack, rule, and admission-policy semantics. \mathsf{Clock} assigns commit time. \mathsf{Status} supplies authenticated snapshots. \mathsf{Store} holds, for each claim, a private state and the non-repeating head \mathit{base}=(\mathit{revision},\mathit{receiptRoot},\beta,d(s)). It also holds a nonce-to-request-digest map. \mathsf{EvidenceStore} retains the full request, proposed facts, resolved semantic objects, assigned time, authenticated snapshot, admitted facts, decision certificate, derivation when present, and receipt by private content digest under its bound retention policy. Section 6.1 states the complete-input condition for replay and the effect of authorized disposal. \mathsf{Eval} computes one pre-receipt decision. \mathsf{Receipt} constructs the canonical receipt.
An ordinary request must name the active binding \beta in the observed base. Changing that binding requires an authorized carriage or economic transformation that commits the old and new bindings and a new base. The trace theorem below applies within a binding. Section 7 extends preservation across meaning-preserving carriage and separates it from amendment.
The family owns claim meaning. The environment owns serialization, content resolution, time, status freshness, and idempotency. A theorem about a family alone cannot establish those engine properties.
3.5 Transition requests, results, and receipts
The request binds the event and the evidence proposed for it. The decision records what the evaluator concluded. The receipt then commits that conclusion together with the inputs and changes needed to audit it. This order avoids letting a receipt’s description determine its own permitted effect.
Definition 3.11 (Transition request).
A transition request is q=(\mathit{claim},h_F,h_T,h_R,h_\Pi,a,\mathit{origin},\mathit{base}, d(\widehat E),\mathit{nonce},\theta,\mathit{submitter},\mathit{signature}). h_R is the content digest of the applicable rule bundle, including every route-specific status-to-evidence-class map. \widehat E is the proposed fact bundle. The origin is \mathsf{Authored} or \mathsf{Derived}(u) for a stable derived identifier submitted by the pack’s authorized scheduler. The submitter asserts event time \theta. The signature covers every preceding field. The engine assigns commit time \tau and computes the admitted subset at commit.
Definition 3.12 (Transition result).
After ingress, the evaluator first computes \eta\in\mathsf{Decision}_{F,T} and its reliance scope. Receipt construction then returns exactly one claim result: \mathsf{Accepted}(s',r),\qquad \mathsf{Refused}(c,r),\qquad \mathsf{Awaiting}(m,r). These are the receipt-bearing images of \mathsf{Accept}(s',\pi,\omega), \mathsf{Refuse}(c,\nu), and \mathsf{Await}(m,\nu), respectively. A refusal code c names a closed reason class. An awaiting requirement m names the missing external fact and the authority that can supply it. An invalid signature, a submitter without standing, a stale base, an unavailable required status check, or conflicting nonce reuse is rejected at ingress. The engine records that rejection as an observation outside the claim trace; it does not change the claim base.
The three results distinguish a permitted change, a known failure, and a question that needs evidence from its named authority. Refusal and awaiting preserve the private claim state. Each novel committed decision still advances the audit revision and receipt root. Thus non-mutation here concerns the private claim state, not the absence of an audit record.
The receipt’s obligation delta is the pair of before and after projections, rather than an untyped numeric difference. This permits a family to expose several dated or contingent obligations through the same receipt structure.
Definition 3.13 (Receipt).
For a novel request nonce, a receipt commits to \begin{split} r=\mathsf{Commit}(&H(\mathsf{Enc}(q)),\mathit{claim},h_F,h_T,h_R,h_\Pi, \theta,\tau,h_\Sigma,\\ &\mathit{requestedBase},\mathit{observedBase},d(E),d(\eta),d(\mathit{cert}(\eta)), d(s),d(s'),\\ &\Delta B,W,\Delta L,\mathit{relianceScope}). \end{split} For non-acceptance, s'=s, \Delta B=(B_F(s),B_F(s)), W=\varnothing, and \Delta L=\varnothing. Its negative certificate remains present and replayable. The full request commitment binds its nonce, asserted event time, proposed facts, content digests, submitter, signature, and requested base.
A receipt is evidence of the engine decision under committed inputs. It is not evidence that an external payment settled, that title passed, or that a court will recognize the result. Those propositions require the external evidence and reliance policy named in the receipt.
The family state records the operational position represented by the engine. It does not decide the legal position. Applicable law and the instrument determine whether an entry is constitutive, evidentiary, or neither. For example, under sections 112 and 127 of the Companies Act 2006, membership outside the initial subscribers requires agreement to become a member and entry in the register, while the register is prima facie evidence of the matters it must contain [7]. Under section 136 of the Law of Property Act 1925, a legal assignment of a debt or other legal thing in action requires an absolute assignment in writing under the assignor’s hand and express written notice to the debtor, trustee, or other person from whom the assignor would have been entitled to claim [18]. An engine receipt commits to and retains evidence of those acts. It cannot supply a missing agreement, signature, entry, or notice. A reliance scope identifies the proposition and audience for which the receipt may be used. It does not decide the proposition’s legal effect.
3.6 Typed diagnostics and evidence cure
The transfer to a holder with missing eligibility evidence needs a precise answer: which assertion is missing, about whom, and from which source. A diagnostic pairs a fixed reason tag with those particular values. A cure supplies evidence for a recorded requirement and triggers a fresh evaluation under the applicable current rules.
A diagnostic separates a finite reason vocabulary from its data. For schema version v, define \mathsf{Reason}_F=\sum_{t\in\mathsf{ReasonTag}_F}\mathsf{Payload}_{v,t}, \qquad \mathsf{Requirement}_F= \sum_{t\in\mathsf{RequirementTag}_F}\mathsf{Payload}_{v,t}. These dependent sums need not be finite. A missing-fact tag can carry a predicate, subject, authority, and evaluation date. The same tag therefore describes different entities and dates without adding codes. The content-addressed schema fixes field types, canonical encoding, and size limits. It rejects unknown fields, ill-typed values, and unsupported schema versions before any claim write. A malformed certificate is a validation error. It supplies neither a refusal nor an awaiting judgment.
The concrete schema uses identifiers of at most 256 UTF-8 bytes, nonnegative integers of at most 256 bits, and canonical ISO calendar dates. A diagnostic occupies at most 4096 bytes. Its finite tags distinguish missing facts, blocked predicates, quantity bounds, consumed allocations, lifecycle conditions, and pending derived events. The payload identifies the exact subject and required source.
Evidence must remain admissible while execution awaits it. The input interface has three constructors: \begin{aligned} &\mathsf{Input}(\widehat E),\\ &\mathsf{Cure}(d(r),m,\widehat E),\\ &\mathsf{Review}(d(r),\mathit{reviewKind}). \end{aligned} Each signed envelope binds the claim, submitter, operation, and input nonce. Input standing permits delivery. It grants no authority to assert the delivered facts. A cure names an existing awaiting receipt and its exact requirement. A review requests a named reconsideration procedure. The resulting decision or evidence enters through its ordinary authenticated source.
Input operations append to a separate immutable journal and proposed-fact store. Their nonce namespace is distinct from claim-transition nonces. An exact retry returns the recorded input result. They change no private claim state, claim revision, receipt root, allocation, funding reservation, or derived disposition. The input service accepts adverse or superseded assertions as historical records. Only admission selects the authoritative, current assertion for a transition. Evidence delivery cannot advance an authority source’s status root.
A recorded relevant input wakes the scheduler. A retry binds an explicit proposed-fact bundle, current base, and fresh transition nonce. The evaluator recomputes all guards, including current authority and prohibitions. A known prohibition therefore remains a refusal after an unrelated cure. An awaiting derived identifier remains pending. A refused identifier remains terminal. An authorized family transition may advance the decision generation after changed facts. It retains the same economic duty, the earlier refusal, and the new event’s authority.
Evidence delivery and claim decisions use separate journals because uploading a fact should not change the obligation. The first proposition states this separation for an execution whose decision inputs remain fixed. The next adds the availability and scheduling premises needed for delivered evidence to resolve pending work.
Proposition 3.14 (Input noninterference).
Deleting input operations from an execution leaves its claim-state transitions and resource writes unchanged, provided each request retains its bound fact bundle and historical source-head inputs.
Proof. Each input operation is the identity on those projections. A claim transition uses its explicit bundle and admitted source snapshot. The retained source history reconstructs the heads used for that decision. Deleting input journal steps while fixing these historical inputs therefore preserves every claim-step input and output. ◻
Proposition 3.15 (Progress after sufficient cure).
Fix a finite closed set of pending derived identifiers with a well-founded order. Suppose each has a finite sufficient evidence bundle. Suppose required facts, authority, and status remain available and valid during retries. Assume terminating evaluation, fair scheduling, starvation-free storage, and eventual input delivery. Every identifier then reaches a terminal decision. An identifier whose complete guards pass is accepted.
Proof. Consider the least pending identifier. Its sufficient bundle eventually arrives and the scheduler eventually retries it. The complete bundle resolves every requirement, including those revealed by earlier evidence. Stable admission and terminating evaluation produce acceptance or refusal. The identifier leaves the pending set. Induction on the finite ordered set proves the assertion. Input delivery does not change a claim base, so competing uploads do not cause base starvation. ◻
The proposition names its environmental progress premises. An unanswered discretionary question remains awaiting until its authorized source answers. It creates no substitute authority and changes no pending obligation.
3.7 Finality strata and lifecycle records
The word “final” has four distinct referents. Engine finality means that the serialized claim trace has committed a result. Asset finality means that the authoritative asset book treats its position entry as irrevocable under its rules. Cash finality means that the designated payment system treats its cash leg as irrevocable and unconditional. Legal settlement means that the instrument, applicable law, and governing rulebook treat the obligation as discharged. A bound rule may map one stratum to another. No mapping is implicit. In particular, engine finality alone implies none of the other three.
A funding instruction, a settlement-ledger confirmation, and credit to the named beneficiary are different records. A claim is described as paid only when authenticated evidence establishes beneficiary credit under its bound rule. For example, Regulation J makes a Fedwire funds transfer final between the relevant participants at the event specified by section 210.31. The receiving bank’s obligation to its beneficiary remains a separate question under Appendix A [5]. Participant finality therefore does not by itself establish beneficiary credit.
Op: A Typed Bytecode for Compliance-Carrying Operations defines the typed external-effect boundary and authenticated provider-finality evidence consumed here [23]. Recourse defines the contested-outcome route and keeps award, execution, satisfaction, and recovery separate [24].
The authoritative asset book can be external to the engine. A central securities depository, registrar, or other named book can establish the asset position under the pack’s adapter and authentication rules. The engine records that evidence. It does not become the authoritative book by observing it.
An institution can expose the distinction through a lifecycle projection J_F(T_{\leq i}) from the committed trace prefix. The projection can give each instruction one stable identifier and an append-only sequence such as \begin{aligned} \mathsf{Received}&\longrightarrow\mathsf{Validated} \longrightarrow\mathsf{Matched}\longrightarrow\mathsf{RiskAccepted}\\ &\longrightarrow\mathsf{SettlementInstructed} \longrightarrow(s_A,s_C), \end{aligned} where each leg is pending, ready, or final. Ready means that every condition for issuing the external settlement instruction is satisfied; it does not mean final. The projection can show \mathsf{LegallySettled} only when both legs are final and the bound instrument, legal rule, and authority evidence establish that consequence. Rejection, an unresolved match, or a curable hold is a typed result. Cancellation, correction, force transfer, and reclamation are new events. None rewrites an earlier event or receipt. The projections B_F and L_F are inputs to J_F but do not determine validation, matching, risk acceptance, or legal settlement by themselves. The lifecycle record is not an independent claim state. Prefix extension can append or refine a current status under the bound temporal rules; it cannot erase an earlier record.
3.8 Corporate actions
Splits, dividends, mergers, rights offerings, tenders, conversions, and redemptions are family events rather than exceptional edits. A corporate-action request must bind its type, declaration, record and effective times, parameters, authority, affected positions and holders in that claim, and requested base. Its family derivation must state exact holder, issued, burnt, cash, escrow, reserve, and obligation deltas. It must also establish the relevant funding, issuance-cap, target-holder, settlement, and lifecycle conditions. This engine commits one atomic effect block only when the action belongs to one claim record. An action spanning claims requires either one composite claim state or an engine extension with a joint base and vector compare-and-swap. The theorem below does not establish that extension. A missing curable condition yields \mathsf{Awaiting}; an invalid action yields \mathsf{Refused}. Neither decision changes claim state. Conflicting actions need an ordering rule in the bound family semantics. Calendar order alone does not supply one.
For example, a cash-dividend activation map fixes the declared amount, record snapshot, eligible positions, payment date, funding route, and rounding rule. It emits one complete effect record per credited position. A split activation map fixes the ratio, record snapshot, treatment of fractional positions, and effective time, then emits the exact old and new quantities. Declaration does not prove funding, and an emitted payment instruction does not prove beneficiary credit. One complete action-effect record has the typed fields \begin{aligned} (&\mathit{actionId},\mathit{positionId},\mathit{holder},\mathit{effectKind}, \mathit{oldQuantity},\mathit{newQuantity},\\ &\mathit{cashAmount},\mathit{currency},\mathit{settlementKey},\mathit{status}), \end{aligned} with an explicit none value for a field that the effect kind does not use. Formally, the partial activation map has type \alpha_F:T_{\leq i}\times\mathcal{S}_F\times\mathsf{CorporateAction}_F \rightharpoonup\mathcal{P}_{\mathrm{fin}}(\mathsf{ActionEffect}_F). It is defined only when the authority, record-time, effective-time, pack, and family guards hold. Both displayed action classes use this map.
Definition 3.16 (Event atomicity).
An accepted event in the base engine is atomic when all engine writes for one claim record are indivisible at one linearization point. This includes its state, nonce, consumed-effect set, receipt, revision, and receipt root. Base-engine atomicity stops at an external settlement rail and at the boundary between claim records. Section 5.2 supplies a separate joint extension. Neither construction asserts simultaneous external cash movement.
The theorem also excludes an outcome-sensitive batch of several instructions. Any such extension must bind the batch membership and order digest before the outcome is revealed. It must then resolve the outcome, execute the already fixed instructions in that order, and reject any post-revelation addition, removal, or resequencing. The derived-event schedule does not supply those batch semantics.
4 Engine and family laws
The example separates two possible errors. A family can calculate an invalid successor, or an engine can commit competing successors that were each valid against an earlier base. Local family laws address the first error; storage, admission, and receipt laws address the second. The trace theorem will compose these obligations across an ordered history.
The preservation result requires both sets of laws. The engine laws are not delegated to the family.
4.1 Engine laws
These are assumptions on a conforming engine, not conclusions about every storage system. The remote protocol below realizes the current-source check under additional cooperation and recovery premises. Concrete family constructors later show how the local transition laws can be discharged.
Assumption 4.1 (Canonical and content-addressed semantics).
\mathsf{Enc} is injective on well-typed values. H is collision resistant. The registry never resolves one digest to two semantic objects. Version labels can be metadata, but receipts and historical audit replay bind the active digests h_F, h_T, h_R, and h_\Pi.
Assumption 4.2 (Commit-time admission).
Immediately before compare-and-swap, the engine assigns \tau, requires \theta\leq\tau, obtains a snapshot \Sigma_\tau that satisfies each predicate’s source-specific freshness bound, and recomputes admission. Every accepted derivation uses only E_{\Pi,x,a,\theta,\tau,\Sigma}. Every fact classified as commit-current is valid and not revoked in \Sigma_\tau. The engine rechecks, at the linearization point, only the source heads used by the request. A changed head forces reevaluation against the new snapshot.
Assumption 4.3 (Authenticated ingress and standing).
The engine verifies the request signature, requires the request’s four semantic digests to equal the active binding in the observed base, and requires the submitter to have standing under that binding before the request can enter the claim trace. A stale base, inactive binding, invalid signature, absent standing, unavailable status source, or conflicting nonce produces an ingress observation without a claim-state write, revision increment, or receipt-root extension.
Assumption 4.4 (Atomic serial extension).
For a novel nonce that passes ingress, reading the observed base, checking the requested base, validating that every source head used in h_\Sigma remains current, computing one authored or derived event decision, consuming the nonce, writing state, appending the receipt, incrementing the revision, and setting \mathit{receiptRoot}'=H(\mathsf{Enc}(\mathit{receiptRoot},d(r))) form one atomic action. A changed source head restarts evaluation. A changed claim base returns an ingress observation. Neither event enters the claim trace. Every claim result that commits advances the receipt root and revision exactly once. A derived event is a separate request and atomic action.
Assumption 4.5 (Nonce idempotency).
The nonce registry maps every committed result, including refusal and awaiting, to the full request digest and receipt. Repeating the same nonce and request digest returns the existing receipt without another append. Reusing the nonce with a different request digest cannot enter the claim trace.
Assumption 4.6 (Deterministic evaluation).
For fixed content-addressed semantics, request, admitted facts, commit time, status snapshot, observed base, and canonical receipt construction, evaluation has one decision. In particular, the family relation is functional in (s',\pi) for fixed (s,a,E,\theta,\tau,\Sigma), and the ordered procedure \mathsf{PackDecision}_F is total and deterministic.
Assumption 4.7 (Evidence retention).
For every committed result, the engine retains the full request preimage, proposed fact bundle, resolved semantic objects, assigned commit time, authenticated snapshot and head attestation, admitted fact set, decision and its certificate, accepted derivation when present, and receipt. A digest authenticates a produced record; it does not replace the record. The policy records each required retention interval and hold. A disposal record identifies every replay range whose input closure it changes.
Assumption 4.8 (Receipt coherence).
For an accepted decision \eta=\mathsf{Accept}(s',\pi,\omega) from s, the receipt records \begin{aligned} \Delta B&=(B_F(s),B_F(s')), & W&=W_F(\pi),\\ \Delta L&=\{k\mapsto(L_F(s)(k),L_F(s')(k)):\\ &L_F(s)(k)\ne L_F(s')(k)\}. \end{aligned} and the digests of \eta and \omega. Its reliance scope is no broader than Z_{F,T}(s,a,E,\eta). For a non-accepted decision, the receipt records the digest of the negative certificate, the observed before and after states are equal, \Delta B=(B_F(s),B_F(s)), W=\varnothing, and \Delta L=\varnothing.
4.2 A protocol for remote commit-current predicates
A source can report valid authority and then change before the claim commits. A separate store can also allocate the proposed payment elsewhere during that interval. Preparation reserves the records a transaction needs while one durable decision is formed. A fence prevents a conflicting write from passing an unresolved preparation. The transaction’s footprint is the complete set of records and predicates it reads or changes.
The commit-current premise has an operational realization when every required source supports authenticated preparation and fences conflicting head updates. The following protocol defines that realization. Each transaction has a stable identifier t. It binds the request, semantic digests, participant set, remote predicate set, and complete read and write footprints. The dependencies include the policy and authority roots that determine the current owners, participants, and footprint. Preparing previously selected participants does not validate a changed participant-selection policy.
Definition 4.9 (Prepared status transaction).
Each source atomically prepares its current relevant heads and returns a signed vote. The vote binds t, the request digest, prepared heads, source epoch, and a deadline in the decision service’s authenticated time. The source retains a durable fence against conflicting head updates. The claim store prepares the requested base and nonce. It reserves the complete write footprint. An allocation store prepares every resource used by the extension in Section 5.2.
The participants name an authenticated, replicated decision service in the bound policy. Its durable decision record has exactly one terminal value, \mathsf{Commit} or \mathsf{Abort}. To commit, the service assigns \tau, checks every prepared vote and deadline, and evaluates the request using the prepared state and status. It verifies all time-dependent predicates at \tau. The evaluation must use only the prepared footprint. An additional dependency or write requires abort and a new preparation.
A commit record binds the complete decision, resulting write vector, prepared votes, and \tau. Its durable creation is the transaction’s linearization point. Deadline validity is checked at that durable decision using the specified authenticated time rule. A proposal timestamp before expiry cannot authorize a decision that becomes durable after expiry. Each participant applies that decision once. Prepared state remains unavailable to conflicting transactions until the decision has been applied. An abort releases reservations without applying claim or allocation writes.
A source that must change a prepared head first resolves each conflicting transaction. If commit wins, the head update follows that commit. If abort wins, the update follows the abort. The source cannot advance the head while leaving an undecided prepared vote usable. The same rule applies to governing policy roots and owner mappings.
Expiry prevents a later commit. It does not authorize unilateral release of a prepared fence. A valid earlier commit may exist behind a failed connection. The participant obtains a terminal decision before releasing the fence. The decision service records abort after expiry if no commit exists. Recovery queries that durable decision record and never infers abort from silence. It reconstructs unresolved preparations and their fences before accepting successor writes. An epoch change cannot discard a preparation that the decision service may already have committed.
Proposition 4.10 (Current heads at a prepared commit).
Assume authenticated votes, durable unique decisions, linearizable preparation, and the stated fencing and application rules. At a committed transaction’s linearization point, every required source head equals its prepared head. Every participant applies the committed state and resource vector at most once.
Proof. Preparation reads and fences each head in one linearizable operation. A conflicting update cannot pass the fence before the terminal decision. Thus every prepared head remains current when the commit decision is recorded. The same argument preserves each prepared claim base and allocation resource. Decision uniqueness gives one terminal write vector. Each participant records application by transaction identifier, so retries apply no second vector. ◻
This protocol lets independent institutions retain their own source write authority. It requires cooperation in the named transaction rather than a shared global institutional ledger. External settlement retains its own provider occurrence and finality evidence.
4.3 Family laws
For the zero-coupon claim, issuance creates the debt, transfer changes the holder, and a final payment reduces the debt. The family must justify each change under the same interface. In particular, every external fact that determines an obligation delta must appear as a keyed effect. Hiding the same payment inside another transformer would evade a consumption check.
Assumption 4.11 (Evaluator soundness).
Every result is the receipt-bearing image of the total decision returned by \mathsf{PackDecision}_F. The evaluator returns an accepted decision only if the request names the claim and active semantic binding, every external fact used belongs to the commit-time admitted set, no higher-priority due derived identifier is pending, and there are witnesses \pi\in\Pi_F and \omega\in\Omega_{F,T} such that U_F(s,a,E,\theta,\tau,\Sigma,s',\pi) \quad\text{and}\quad \mathsf{PackAdmit}_F(T_F,s,a,E,\theta,\tau,\Sigma,\pi,\omega)=1. A refused or awaiting decision carries the canonical negative certificate for the first failed or missing condition.
Assumption 4.12 (Authority).
Every accepted derivation satisfies A_F(\pi)=1 under its bound event time, commit time, and status snapshot.
This law says that every accepted event has the authority witness required by the family rule. It does not prove that the family rule correctly states applicable law.
Assumption 4.13 (Invariant preservation).
If P_F(s)=1 and U_F(s,a,E,\theta,\tau,\Sigma,s',\pi), then P_F(s')=1.
Assumption 4.14 (Obligation accounting).
For every accepted derivation \pi, B_F(s')=D_F(\pi)(B_F(s)). The derivation states whether the transformer creates, amends, transfers, discharges, disputes, recovers, or closes obligations.
Assumption 4.15 (Keyed effect allocation and single consumption).
Every effect premise in \pi supplies exactly one use u\in V_F(\pi) and carries an issuer-authenticated occurrence key that \kappa_F maps to \mathsf{key}_F(u). Every use is backed by such a premise. The map \mathsf{key}_F is injective on V_F(\pi), and \operatorname{dom}(\Delta_F^{\mathrm{eff}}(\pi))=W_F(\pi) =\mathsf{key}_F[V_F(\pi)]. Thus one accepted derivation has one total effect delta for each key and applies that delta once in the canonical fold defining D_F(\pi). The delta is the complete family effect of that occurrence, not an aggregation of repeated uses. Every admitted fact whose occurrence or attributes determine an obligation delta is an effect premise. Status premises can guard a fixed transition, and settlement-observation premises can update L_F, but neither can parameterize D_F^0(\pi) or an effect delta. The only dependence of D_F(\pi) on effect premises is through these keyed deltas. An unmapped premise makes the derivation ill formed. D_F^0(\pi) is independent of every effect premise. If the transition is accepted, then W_F(\pi)\cap C_F(s)=\varnothing \quad\text{and}\quad C_F(s')=C_F(s)\cup W_F(\pi). This is a family proof obligation: the engine checks the typed allocation but cannot infer it from a private transformer.
Assumption 4.16 (Issuer key discipline).
The admission policy gives every admitted external occurrence one canonical effect key across all permitted issuer namespaces and routes. It records the authoritative namespace and every admitted alias. Every authenticated assertion about the occurrence resolves to that key. If the policy cannot establish that two identifiers are distinct or are aliases, neither can support an obligation effect. Claim, amount, currency, and status are authenticated attributes; changing one does not create a new occurrence.
Assumption 4.17 (Holder admissibility).
If G_F(s')\neq G_F(s), then every p\in G_F(s')\setminus G_F(s) satisfies H_F(\pi,p)=1 under the bound rule content, times, and status snapshot.
Assumption 4.18 (Effect-indexed finality non-escalation).
For each k\in\mathsf{SettlementKey}_F, Q_F(\pi)(k) is computed only from admitted, issuer-authenticated settlement-effect or settlement-observation premises mapped to that key. It is \mathsf{None}_{F,k} when \pi has no such premise. For every key k whose settlement entry changes under \pi, L_F(s')(k)\preceq_{F,k} Q_F(\pi)(k). A settlement-observation premise may therefore change L_F with W_F(\pi)=\varnothing; it cannot change B_F through an effect delta. Except for a correction that the bound route permits before finality, each changed entry also satisfies L_F(s)(k)\preceq_{F,k} L_F(s')(k). A consumed settlement key is at \mathsf{Final}_{F,k}. Once an entry reaches \mathsf{Final}_{F,k} it remains there; a reversal is a separately keyed occurrence. An unmodified key keeps its old entry. The receipt records before and after values for exactly the changed keys. Evidence for one payment cannot promote another payment or the whole claim.
Assumption 4.19 (Reliance non-escalation).
Every party class, proposition, jurisdiction, and expiry in Z_{F,T}(s,a,E,\eta) has a named rule or admitted reliance-authority premise. With no such premise, Z_{F,T} permits only engine-audit reliance on the proposition that the engine produced the receipt. The receipt scope satisfies \mathit{relianceScope}(r)\sqsubseteq_F Z_{F,T}(s,a,E,\eta). The permitted scope is itself derived from bound rules and admitted external evidence. A family cannot obtain a broader scope by naming it in the receipt.
Assumption 4.20 (Non-mutation on non-acceptance).
A refused or awaiting request leaves the observed private claim state unchanged. Its receipt can advance the audit root and revision, but it records no claim-state, obligation, effect, or settlement delta.
These laws form the interface. A new family enters the common lifecycle only after it supplies a concrete derivation system and proves the family laws. A conforming engine separately establishes its registry, status-freshness, evidence-retention, nonce, serialization, and receipt obligations. A product label is insufficient.
5 Parametric lifecycle preservation
A valid trace is the history obtained by repeatedly committing the specified single-step operation. It includes refusals and awaiting decisions because they have receipts, although they leave the private obligation unchanged. The theorem lifts the local invariant law to that history. Its separate consumption argument uses the retained set of earlier effects; its base argument uses the strictly increasing revision.
Definition 5.1 (Serialized valid trace).
For claim instance x=(\mathit{id},F,\beta,s_0) in environment \mathcal G, a trace is T_x=(s_0,b_0,q_1,\eta_1,r_1,s_1,b_1,\ldots,q_n,\eta_n,r_n,s_n,b_n), where each \eta_i\in\mathsf{Decision}_{F,T} and b_i=(\nu_0+i,\rho_i,\beta,d(s_i)) for the initial revision \nu_0; the active binding \beta is fixed. It is valid when each novel request is processed by the engine’s atomic action. Every accepted decision has requested base b_{i-1}, a derivation \pi_i, and a pack-admission witness \omega_i satisfying \begin{gathered} U_F(s_{i-1},a_i,E_i,\theta_i,\tau_i,\Sigma_i,s_i,\pi_i),\\ \mathsf{PackAdmit}_F(T_F,s_{i-1},a_i,E_i,\theta_i,\tau_i, \Sigma_i,\pi_i,\omega_i)=1. \end{gathered} Every refused or awaiting decision has s_i=s_{i-1}. A derived identifier’s disposition is updated from its decision: awaiting remains pending, while acceptance and refusal are terminal. Every step sets \rho_i=H(\mathsf{Enc}(\rho_{i-1},d(r_i))). An exact idempotent retry and an ingress observation are not new trace steps.
Theorem 5.2 (Engine-relative parametric lifecycle preservation).
Let F satisfy the family laws, including issuer key discipline. Let \mathcal G satisfy the engine laws. Let T_x be a serialized valid trace with P_F(s_0)=1. Then:
- (P1)
-
P_F(s_i)=1 for every reached private state s_i.
- (P2)
-
Within an accepted step, no two members of V_F(\pi_i) have the same effect key. No effect key belongs to W_F(\pi_i) at two distinct accepted steps. Thus one external occurrence is consumed at most once in the claim trace.
- (P3)
- The committed bases form a strict non-repeating chain: the revision of b_i is \nu_0+i, its active binding is \beta, and \rho_i=H(\mathsf{Enc}(\rho_{i-1},d(r_i))). Every accepted step extends the observed current base. Refused and awaiting results extend the receipt chain while leaving private claim state unchanged.
Proof. Proceed by induction on committed trace length. The initial state satisfies (P1) by hypothesis. A refused or awaiting result leaves the private state and consumed-key set unchanged. An accepted result has a family derivation by evaluator soundness. Family invariant preservation gives (P1).
For (P2), \mathsf{key}_F is injective on V_F(\pi_i), so two effect uses in one accepted step cannot consume the same occurrence. The domain equality gives one total delta and one application for each such key. Across steps, keyed effect allocation requires W_F(\pi_i)\cap C_F(s_{i-1})=\varnothing and then adds every member of W_F(\pi_i) to C_F(s_i). The consumed-key set never shrinks. A key used at step i therefore cannot occur in the support of a later accepted step. Issuer key discipline makes this a statement about external occurrences, not only identifiers.
Atomic serial extension increments the revision once and hashes the previous receipt root with the new receipt. The incremented revision makes the base strictly new. Injective encoding and collision resistance authenticate the history committed by its receipt root; they are not needed to show that the base does not repeat. An accepted step must name the observed base. Non-acceptance leaves s_i=s_{i-1}. These facts prove (P3). ◻
Remark 5.3 (What the theorem does not prove).
The theorem is conditional on the engine and family laws. It does not establish the atomicity of a particular store, the honesty of a clock or status source, the validity of an authority document, the legal adequacy of a holder rule, or the truth of a provider assertion. Receipt coherence, holder admissibility, obligation accounting, settlement-evidence monotonicity, and reliance non-escalation remain explicit interface laws rather than additional theorem claims.
5.1 Disjoint claims and the boundary of netting
The single-claim result allows concurrent work only when its dependencies permit it. Separate claims can still compete for one payment, cash balance, or authority record. The next two propositions give conditions under which changing the execution order leaves both results unchanged.
Proposition 5.4 (Disjoint-claim non-interference).
Let accepted requests q_1 and q_2 operate on distinct claim records. Assume that every mutable engine write location used by one request is disjoint from every such location used by the other, including nonce entries, claim states, receipt roots, and consumed-effect registries. Assume also that their canonical effect-key supports are disjoint, neither transition changes a status head read by the other, and their assigned commit times and authenticated snapshots are fixed. Then the two atomic extensions commute on the product of the two claim records. Each order reaches the same private states, obligation projections, settlement ledgers, consumed-key sets, revisions, and per-claim receipt roots. Both family invariants remain true.
Proof. Each extension reads immutable registry content and its fixed snapshot. By hypothesis, every mutable location and occurrence key used by one extension is disjoint from those used by the other. Applying the first extension therefore changes no input to the second. The componentwise updates commute, and the single-claim preservation theorem applies to each projection. ◻
A shared ordered constraint permits a second case. Each request tightens the constraint by taking a meet, the greatest value below both inputs. If the combined tightening still satisfies both requests’ thresholds, their shared updates can commute. The hypotheses also exclude hidden dependence on the intermediate value.
Proposition 5.5 (Compatible shared-coordinate non-interference).
The conclusion also holds when the requests share only coordinates z in a meet-semilattice. For each shared coordinate, request j tests t_j\leq z and writes z\leftarrow z\wedge w_j, where t_j and w_j are fixed by the request. Each request reevaluates that test against the current z at its linearization point. Apart from the truth of this test, its decision, certificate, private writes, and receipt are independent of the value of z. Assume t_j\leq z\wedge w_1\wedge w_2\qquad(j=1,2), so the own and cross-request thresholds both survive. Assume all other mutable writes and effect keys are disjoint and the status heads are fixed. Then the two extensions commute and preserve both threshold conditions.
Proof. Meet is associative, commutative, and idempotent, so either order writes z\wedge w_1\wedge w_2. The displayed inequalities preserve both tests. Every other component commutes by disjointness, and the single-claim theorem applies to each claim projection. For sorted finite write footprints and constant-time lattice comparison, checking these conditions is linear in the combined footprint size. The threshold form is the same composition discipline used for compliance grades [20]. ◻
The proposition is request-specific non-interference, not a closure theorem for families. A product family would also need definitions for initial state, derived-event priority, transition and pack-admission witnesses, obligation and holder projections, authority, tagged occurrence keys, settlement orders, reliance, and combined failures. Each family law would then need a proof for the product. Disjointness of one pair of requests does not supply those definitions. A composite instrument also needs a pack recognized by a named regime for the relevant purpose.
Netting does not satisfy these hypotheses merely because its arithmetic sums to zero. For claims 1,\ldots,m, a netting extension needs the joint obligation vector \mathbf b=(B_{F_1}(s_1),\ldots,B_{F_m}(s_m)) \in\prod_{j=1}^m\mathcal O_{F_j}. The pack must also name a commutative exposure monoid (A,+,0) and typed maps e_j:\mathcal O_{F_j}\to A. For every admitted key k, it names a native effect type \mathsf{EffectDelta}(k) and a typed normalization \iota_k:\mathsf{EffectDelta}(k)\to A. The joint extension binds a total delta \Delta_k\in\mathsf{EffectDelta}(k). Its derivation must bind the included claims, legal basis, authority, pre-net vector \mathbf b, post-net vector \mathbf b', and an allocation (a_{k,1},\ldots,a_{k,m})\in A^m for every canonical effect key k. Here a_{k,j}=0 when k does not change component j, and \sum_j a_{k,j}=\iota_k(\Delta_k). The joint transformer must prove \mathbf b'=N(\mathbf b) under that allocation. The pack must name the conserved exposure projection and prove its exact pre-net and post-net equation; any settlement delta must equal the sum of its allocated final effects. One atomic cross-claim registry must consume each allocated key once. Close-out netting or novation changes obligations under its contract and legal authority. Cash settlement changes them only under admitted final-settlement evidence. These are minimum interface requirements, not a general netting theorem. The pack’s \mathsf{NettingBasis} identifies the asserted legal model. It does not discharge these joint-state obligations.
AMM-Aware Central Counterparty Risk develops multi-claim clearing and risk allocations. This paper supplies the occurrence-attribution and funding contracts for their joint execution under the pack’s declared netting basis [25].
5.2 Occurrence attribution and funding
A completed payment and a current cash balance are different resources. Attributing a receipt determines which obligations that occurrence satisfies. Reserving current cash determines whether another command can spend it. The engine represents both resources and commits their applicable constraints with the affected claim transitions.
Suppose one final credit of 100 supports two claims. Allocating 60 to the first must leave at most 40 for the second. Stable portions record exactly which parts each claim used. A later correction or reassignment must retain those portions’ ownership history, so that restored evidence cannot silently fund another claim.
Definition 5.6 (Canonical occurrence and attribution record).
The admission policy nominates one authoritative booking for each occurrence o. Its key contains the authoritative namespace, record kind, and stable event identifier. The policy checks the account, asset, book, and identifier epoch that give the key its scope. Reporting wrappers resolve through authenticated aliases. A causal predecessor, interbank posting, or foreign-exchange leg is distinct unless evidence establishes that it identifies the same booking.
The record retains every quantity assertion and its provenance. Each admitted evidence revision e has an immutable authenticated quantity Q_o^e and unit. An amount correction creates a new evidence revision for o. It creates neither a new physical occurrence nor an additional receipt.
The engine divides quantity into stable slices in the declared precision. A slice has identity (o,j) for a positive integer j. An implementation can represent adjacent slices as ranges. Each allocation binds its exact slices rather than only their total amount. The admitted revision fixes a supported slice set V_o^e with |V_o^e|=Q_o^e. Its ordering and correction rule belong to the bound policy. Slice ownership and consumption provenance persist across evidence revisions.
At a reconciled revision, the attribution state contains unallocated quantity U_o, attribution reservations and reconciliation holds R_o, and supported allocations S_{o,c}. The index c names an exact claim and legal allocation purpose. These quantities count disjoint subsets of V_o^e and satisfy U_o+R_o+\sum_c S_{o,c}=Q_o^e, \qquad U_o,R_o,S_{o,c}\geq0. Amounts are integers in the declared precision. The record also retains disputed attributions and all historical allocation entries. Disputed amounts are excluded from supported settlement and available attribution capacity.
Each allocation has an immutable identifier subordinate to o. It binds the claim, purpose, exact slice set, quantity, authorizing transaction, and predecessor allocations. A consumed slice retains its allocation provenance after support is disputed. Changing the reporter, legal origin, command, or observation cannot create another Q_o^e. Replaying an allocation identifier returns its recorded disposition. It does not change the partition again.
Definition 5.7 (Joint attribution transition).
An attribution reservation moves a quantity from U_o to R_o. Acceptance moves its quantity from R_o to the named S_{o,c} and applies the matching family obligation delta. Reservation records distinguish never-consumed attribution slices from reconciliation holds on previously consumed slices. Only a never-consumed reservation returns to U_o after a conclusive abort. A consumed ownership hold resolves through authorized reconciliation and the affected claim adjustments. Rejection or timeout cannot return it to U_o. These changes commit atomically with claim bases, consumed-use identifiers, nonces, and receipts.
An authorized reassignment transfers quantity between allocation owners. It preserves the old journal and records immutable successor identifiers. The joint family derivation adjusts both claims under the governing rules. It cannot retain the source discharge while creating the same discharge at the destination. If the rules retain the source consequence, the destination requires another supported occurrence or an explicit, separately funded obligation.
The consumable identity changes at this extension. The base theorem assigns one whole occurrence to one claim. Partial attribution instead gives each allocated portion its own single-use identifier, while all such identifiers retain the same physical occurrence as their source.
For a monetary settlement effect in this extension, the consumable key is an allocation identifier a. Its evidence package contains the occurrence evidence and the allocation-service witness. The family uses the partial map \kappa_F^+:(\mathsf{OccurrenceKey}\times\mathsf{AllocationId}) \rightharpoonup\mathsf{AllocationId}. It returns a only when the witness binds a to occurrence o, this claim, and the exact supported quantity. This map replaces the occurrence-only canonicalization for allocated effects. The consumed set C_F^+(s) contains allocation identifiers and never shrinks. Nonmonetary effect keys occupy a disjoint tagged namespace. A shared monetary occurrence cannot also supply an unallocated native effect.
The physical occurrence remains separately addressable. Settlement observations are indexed by that occurrence. The family’s allocation-indexed evidence projection follows the bound root and can reach no class above its authenticated evidence. Its obligation delta cannot exceed the supported allocation quantity. The adapter proves the other family laws for this extended state and witness type. Naming an allocation field does not discharge those laws.
The local single-consumption law applies to allocation identifiers. Issuer key discipline supplies one canonical physical occurrence. Allocation-service discipline supplies one canonical identifier for each use. The original single-claim theorem is the special case that assigns one complete occurrence to one claim. The extension admits several partial uses without treating them as several payments.
The partition equation now gives the bound on supported claim credits. Its preservation requires joint claim and allocation writes. Otherwise one store could discharge a claim while another still treats its payment portions as available.
Proposition 5.8 (Attribution conservation).
Fix an admitted occurrence identity and reconciled quantity revision. Assume every attribution transition satisfies the stated partition and commits atomically with its affected claims. Assume evidence reconciliation retains slice ownership under the restoration rule below. Then \sum_c S_{o,c}\leq Q_o^e-R_o. Concurrent requests and replay cannot consume one reservation twice. Reassignment cannot duplicate the corresponding claim credit. A restored supported slice cannot become fresh capacity while retaining its previous claim attribution.
Proof. The initial state places the quantity in U_o. Reservation, acceptance, abort, and reassignment transfer equal quantities between partition members. Induction preserves the equality and nonnegativity. Removing U_o\geq0 gives the displayed inequality. Atomic reservation consumption and immutable use identifiers prevent a second application. Atomic claim updates preserve the stated reassignment condition. Changing evidence support leaves slice ownership unchanged. A restored consumed slice therefore supports its existing allocation. It cannot simultaneously enter the unallocated subset. ◻
Identity and quantity reconciliation.
Aliases resolve to the nominated booking before supported allocation. An unresolved identity cannot independently mint another attribution budget. If later evidence identifies two admitted roots as one occurrence, the engine neither adds their budgets nor discards either history. It freezes further attribution and atomically reconciles identity, quantity, reservations, journals, and affected claim projections. The reconciliation maps old slice references to the canonical occurrence’s slices and retains their consumption provenance. Two claimed uses of one canonical slice cannot both receive support. An unresolved ownership conflict holds that slice in R_o rather than U_o.
A consumed slice outside V_o^e remains attached to its recorded allocation as disputed support. If a later revision restores that slice to V_o^e, it first restores support for that same allocation. It does not make the slice unallocated. An evidence correction alone never changes the recorded allocation owner. A different owner requires an authorized reassignment and the corresponding joint claim changes.
The reconciliation retains each historical attribution A_{o,c} as supported quantity S_{o,c} and disputed quantity D_{o,c}: A_{o,c}=S_{o,c}+D_{o,c},\qquad D_{o,c}\geq0. Supported allocations satisfy the new quantity partition. Here A_{o,c} is the currently effective attribution recorded by the retained journal, not the sum of every historical transfer entry. A prior reassignment remains visible without being counted as another credit.
The bound family and route rules determine supported attribution order and any correction or recovery obligations. If that determination needs evidence, the affected allocation remains explicitly disputed. It supplies no additional spendable or settlement capacity. A failed reconciliation retains the evidence and the frozen state. It cannot release reservations while a previous commit remains possible.
Two histories that each attributed 100 to the same actual 100 receipt expose 100 of unsupported attribution and retain both histories. An assertion correction from 100 to 90 exposes a 10 discrepancy if 100 was attributed. It does not invent a 10 debit. If a later correction restores 100, the same disputed 10 regains support for its existing allocation. It cannot fund another claim. A quantity increase updates the same occurrence’s partition under its authenticated correction rule. Only supported slices without a prior allocation or reservation can enter U_o. It does not mint a second occurrence.
The conservation proposition governs a reconciled evidence revision. Identity or quantity corrections invoke reconciliation rather than rewriting an earlier revision’s quantity or proof inputs. A correction cannot retroactively make erroneous source evidence true. Historical claim consequences remain recorded. Current settlement support reflects the admitted correction and governing rules.
Current cash and command reservations.
A funding domain d identifies its controlling book, account, asset, identifier epoch, and encumbrance scope. Its evidence states the current controlled balance B_d. Its registry contains free capacity F_d, command reservations C_d, encumbrances E_d, and funding deficit D_d: \begin{aligned} B_d+D_d&=F_d+C_d+E_d,\\ F_d&=\max\{B_d-C_d-E_d,0\},\\ D_d&=\max\{C_d+E_d-B_d,0\}. \end{aligned} The balance and encumbrance definitions identify provider holds already included, so the same hold is not deducted twice. Each new funded command reserves its quantity in the applicable domain. Its claim, risk, and collateral constraints share the same atomic footprint. A domain cannot be partitioned merely by naming different claims, products, legal origins, or services.
Attributing a completed receipt changes its attribution partition and matching obligation. It does not debit B_d again. A recipient can spend a received 100 and later attribute that receipt to the original debt. Conversely, unallocated receipt evidence does not prove that 100 remains available for another payment.
An unresolved external command retains its cash reservation. An authenticated debit reconciles that reservation once. An unexplained balance change creates a reconciliation state and any resulting deficit. It does not release reservations because a local command timed out. An authoritative balance change enters the evidence record even when recovery is required.
Returns and reversals.
A refund command reserves current cash. Its completed debit is a distinct occurrence with its own canonical key and single-use attribution. The original credit, quantity assertions, and credit allocations remain in history. The family assigns the debit to its correction or recovery consequence. Net recovery derives from allocated credit and debit occurrences. Ordinary spending is not a reversal of the original debt payment.
An unsolicited return or reversal uses the same debit-occurrence interface. It changes current cash evidence and the applicable recovery projection. It does not rewrite the original receipt or create available attribution capacity. Whether it restores an earlier obligation or creates another is a family rule supported by the governing terms. An amount assertion correction uses reconciliation, not the physical-debit interface.
Joint implementation contract.
The atomic footprint contains each affected occurrence and evidence revision, allocation, funding domain, claim base, authority dependency, consumed-use record, nonce, and receipt. It also contains the policy roots that determine the footprint and its owners. One store can implement the footprint as one serializable transaction. Independent stores require the prepared decision and recovery protocol in Section 4.2. A local claim compare-and-swap alone does not implement the joint action.
Disjoint resources can execute concurrently. Conflicting allocations and funding reservations serialize on their shared domains. Family adapters prove that private obligation deltas match the allocation vector. Provider adapters establish identity, quantity, aliases, balance scope, and finality under the named route. Offline replay reconstructs attribution and funding histories without dispatching another external command.
5.3 The zero-coupon family
The opening example instantiates the transition-critical fragment used in the trace theorem. It does not supply a complete registered pack, derived-event schedule, reliance order, or reason algebra. Fix a registered pack whose admission witnesses establish the authority, holder, settlement, reporting, dispute, and reliance conditions used below. The state is either \mathsf{Draft}, \mathsf{Outstanding}(I,h,100,M,L,C), or \mathsf{Discharged}(I,h,100,M,L,C). Fix a precision c and let \mathsf{Amount}_c\subset\mathbb Q_{\geq0} be the representable amounts at that precision. Here M:\mathsf{SettlementKey}_F\rightharpoonup\mathsf{Amount}_c is a finite partial map from final-settlement keys to amounts, L is the settlement-effect ledger, and C=\operatorname{dom}(M) is the consumed-key set; this family has no other consuming effect keys. Assume 100\in\mathsf{Amount}_c, set \mathsf{EffectDelta}_F=\mathsf{Amount}_c, and define B_F(s)=100-\sum_{k\in\operatorname{dom}(M)}M(k), \qquad G_F(s)=\{h\} for an issued state. Define B_F(\mathsf{Draft})=0, G_F(\mathsf{Draft})=\varnothing, C_F(\mathsf{Draft})=\varnothing, and P_F(\mathsf{Draft})=1. The settlement map M belongs to issued states. The issued-state invariant is 0\leq\sum_k M(k)\leq100,\quad C=\operatorname{dom}(M), and every k\in C is bound by admitted provider evidence to this claim, route, currency, and amount M(k).
The event alphabet contains \mathsf{Issue}(A), \mathsf{Transfer}(h'), and \mathsf{Observe}(k,z,x), where z is a route-defined evidence class; in this example \mathsf{Final} abbreviates \mathsf{Final}_{F,k} for the displayed key. The accepted issuance transition starts at \mathsf{Draft} and requires a current issuer-authority fact. It creates the 100-unit obligation with empty M and C, and with L at its default \mathsf{None} entries. It has no external-effect premise: V_F(\pi)=\varnothing and D_F^0(\pi) is the constant map b\mapsto100. Transfer requires authority from the current holder and an admitted eligibility fact for the destination. It has empty effect use, D_F^0(\pi) is the identity, and it changes G_F but leaves B_F, M, L, and C unchanged.
For \mathsf{Observe}(k,z,x), require k\in\mathsf{SettlementKey}_F. Admitted provider evidence must authenticate k=(\mathit{authoritativeNamespace},\mathit{recordKind}, \mathit{stableEventId}) through the nominated recipient-credit record. A provider transaction reference can be an authenticated alias. Its account, asset, book, and identifier epoch must satisfy the admission policy. A related interbank posting or foreign-exchange leg is distinct, not an alias. The evidence must bind the claim, amount x, and currency as attributes. The derivation records Q_F(\pi)(k)=z. If z\prec_{F,k}\mathsf{Final}, it can change L(k) under the route’s temporal state machine but has the identity D_F^0(\pi) and V_F(\pi)=\varnothing. If z=\mathsf{Final}, it requires k\notin C and 0<x\leq B_F(s). It supplies one effect use u with \mathsf{key}_F(u)=k, sets \Delta_F^{\mathrm{eff}}(\pi)(k)=x, adds M(k)=x, and uses R_F(x)(b)=b-x with identity D_F^0(\pi). The state becomes discharged exactly when the result is zero. A final amount above the outstanding amount is retained in the evidence store and returns \mathsf{Refused}(\mathsf{AmountExceedsOutstanding},r) for this claim. An attempt to finalize a consumed key returns \mathsf{Refused}(\mathsf{EffectAlreadyConsumed},r). A corrected assertion about a known key updates its attributes under the route rule; it cannot create a second occurrence. Before consumption, a correction may replace or lower a sub-final ledger entry only under the bound route’s temporal state machine. An assertion that no longer binds this claim, route, and currency cannot later promote the entry. If the correction changes a consumed amount or finality classification, it cannot rewrite the earlier discharge. The route must supply a distinct corrective or reversal occurrence, or the request remains non-accepted.
A later return, reversal, or insolvency avoidance does not reinterpret the original \mathsf{Final} occurrence. It is a new authenticated occurrence with its own key. The governing terms determine whether that occurrence creates a new obligation, which then enters this family through a new admissible transition or enters another claim family.
On a shared settlement rail, this example uses the full-allocation special case of Section 5.2. The allocation service binds the use of k to one immutable allocation identifier for the exact claim and amount. It reserves that attribution and commits consumption with the claim transition. The private consumed set alone does not establish this cross-claim condition. General partial allocations use allocation identifiers as monetary effect keys while retaining k as the physical occurrence. Attributing a completed receipt does not debit current cash again. A later return uses its own debit occurrence and governing recovery transition.
The settlement ledger is a finite map with \mathsf{None} as its default. The permitted reliance scope always includes the proposition “this engine evaluated this request” for engine auditors. A provider-settlement proposition is available to a claim party only when the admission policy explicitly permits that proposition, party class, route, jurisdiction, and expiry. No zero-coupon rule creates third-party legal reliance.
The eight transitions map to the interface as follows. k_{60} and k_{40} are different provider effect keys. \Delta L shows the old and new value for each changed entry.
| Event | Required premise | W_F | \Delta L | Result and obligation |
|---|---|---|---|---|
| Issue to A | Current issuer authority | \varnothing | \varnothing | Accepted; B=100 |
| Transfer to H | Current-holder authority; H eligible | \varnothing | \varnothing | Accepted; B=100 |
| Transfer to Y | Eligibility evidence missing | \varnothing | \varnothing | Awaiting; B=100 |
| Transfer to X | X is ineligible | \varnothing | \varnothing | Refused; B=100 |
| Final p_{60} | Provider binds k_{60} to 60 | \{k_{60}\} | k_{60}:\mathsf{None}\to\mathsf{Final} | Accepted; B=40 |
| Reuse p_{60} | k_{60}\in C_F(s) | \varnothing | \varnothing | Refused; B=40 |
| Pending p_{40} | Provider binds k_{40} to 40 | \varnothing | k_{40}:\mathsf{None}\to\mathsf{ProviderAccepted} | Accepted; B=40 |
| Final p_{40} | Provider binds k_{40} to 40 | \{k_{40}\} | k_{40}:\mathsf{ProviderAccepted}\to\mathsf{Final} | Accepted; B=0 |
These cases define U_F. The displayed authority and holder premises define A_F and H_F(\pi,p). The two transformers, consumed-key updates, and per-key evidence ceilings prove obligation accounting, keyed effect allocation, single consumption, holder admissibility, and finality non-escalation by case analysis. Refusal and awaiting cases are non-mutating. The family invariant follows because a final event requires 0<x\leq B_F(s).
Corollary 5.9 (Trace conservation and keyed settlement).
In any serialized valid trace of this zero-coupon family, every reached state satisfies 0\leq B_F(s)\leq100. The draft state has B_F(\mathsf{Draft})=0 and C_F(\mathsf{Draft})=\varnothing. Every issued \mathsf{Outstanding} or \mathsf{Discharged} state satisfies B_F(s)+\sum_{k\in\operatorname{dom}(M)}M(k)=100. Each entry of its M is supported by one admitted canonical \mathsf{Final} occurrence for this claim, route, amount, and currency. No canonical provider occurrence supports two entries.
Proof. The draft projections give zero obligation and empty consumption. Issuance creates obligation 100 and an empty settlement map, establishing the issued-state equation. Transfer and sub-final observation preserve that equation. A final observation adds x to the settlement map and subtracts x from the outstanding obligation. Its guard 0<x\leq B_F(s) preserves the bounds. Refusal and awaiting preserve the private state. Induction therefore gives the stated draft and issued-state conclusions. Keyed effect allocation and the settlement rule supply each entry’s canonical key. Issuer key discipline and Theorem 5.2(P2) give uniqueness. ◻
Corollary 5.10 (No duplicate discharge evidence).
Under issuer key discipline, no canonical provider effect key contributes to more than one reduction of the outstanding zero-coupon obligation, even if the provider issues multiple assertions about that occurrence.
Proof. The final-settlement case requires k\notin C_F(s) and then adds k. Within that step the injective use map permits one use of k. Across later steps the consumed-key set grows monotonically. Theorem 5.2(P2) therefore excludes duplicate discharge even when the provider emits several assertions. ◻
5.4 Two concrete family constructors
The zero-coupon example has only an unconditional payable amount. To test the same interface on a different obligation, consider a grant whose payment becomes due after a milestone. Completion must move quantity from contingent to payable, while payment moves it from payable to settled. Assignment leaves both quantities intact. These changes determine the two-component state below.
A closed interpreter instantiates the interface for transferable receivables and milestone grants. A receivable starts with one payable obligation. A grant starts with contingent rights and makes each payable after authenticated completion. Both families support assignment, partial payment, final payment, and historical replay.
The parameter record fixes issuer, initial holder, asset unit, positive integer quantities, component dates, authorities, settlement policy, and reliance scope. It contains a constructor name and declarative fields, with no supplied callback. Schema validation rejects unknown fields, ambiguous identities, nonpositive quantities, and inconsistent units or due order. The family digest binds both constructor semantics and parameters. The reference implementation bounds component sets to 1024 entries. The following rules apply to every finite admitted parameter set.
States and obligations.
Let J be the component set and let q_j>0 be its fixed quantities. A receivable has J=\{\mathit{principal}\}. An issued grant has a finite ordered milestone set. The common state is s=(\iota,h,V,K,M,L). The issuance flag is \iota. The holder is h after issue and absent in draft. The vested set is V\subseteq J. The map K assigns each consumed completion key to its vested component. The map M assigns each consumed settlement allocation to a component and quantity. The evidence ledger L records pending or final evidence for each allocation. The draft has \iota=0 and empty holder, sets, and maps.
In the formulas below, \mathbf1_{j\in V} is the indicator that component j belongs to the vested set. The vector e_j has one at component j and zero elsewhere.
Write p_j=\sum_{a:M(a)=(j,x)}x. The obligation carrier is \mathbb Z^J\times\mathbb Z^J, with projection B(s)=(u,v),\qquad u_j=q_j(\iota-\mathbf1_{j\in V}),\qquad v_j=q_j\mathbf1_{j\in V}-p_j. The first component is contingent quantity. The second is payable quantity. An issued receivable has V=J and zero contingent quantity. An issued grant initially has V=\varnothing. The invariant requires 0\leq p_j\leq q_j\mathbf1_{j\in V},\qquad u_j,v_j\geq0, unique consumed keys, and exact final evidence for every entry of M. Completion and settlement keys occupy distinct typed namespaces. For the grant, K has distinct values and image V. The consumed set is \operatorname{dom}(K)\uplus\operatorname{dom}(M). The holder projection is empty in draft and \{h\} after issue.
Accepted rules.
Issuance requires draft state, the fixed issuer’s signed request, current issuance authority, and initial-holder eligibility. It sets \iota=1 and h=h_0. For a receivable it sets V=J and B=(0,q). For a grant it sets V=\varnothing and B=(q,0). Its effect support is empty and D^0 is the corresponding constant transformer.
Assignment requires issue, the current holder’s signed request, current transferor authority, and destination eligibility. It changes only the holder and retains the admitted assignment evidence. The obligation transformer is the identity and the effect support is empty. The fixed pack determines the legal acts that make assignment effective.
Vesting means that the grant’s stated condition has been met, making the component payable.
A grant’s vesting event names one unvested component j. It requires the issuer’s request and the named completion authority’s current assertion about this grant and component. The completion occurrence has one fresh key k. The successor records K(k)=j and adds j to V. Its single effect delta is \Delta^{\mathrm{eff}}(k)=\mathsf{Vest}(j,q_j),\qquad R(\mathsf{Vest}(j,q_j))(u,v)=(u-q_je_j,v+q_je_j). Its effect-independent transformer is the identity. Completion changes no settlement entry. A distinct component requires its own named completion occurrence.
An observation names allocation a, component j, and exact evidence. It requires an issued, vested component and reporting standing under the pack. The evidence binds claim, component, policy digest, asset, unit, route, purpose, quantity, authoritative occurrence, and allocation slices. The signed claim-allocation predicate certifies the pack’s occurrence-time payee rule. Its named evidence adapter establishes the recipient and assignment facts. The checker validates that certification’s claim, policy, and source binding. A later holder may therefore reconcile an earlier authorized payment.
Pending evidence changes only L(a) and leaves all obligations unchanged. Final evidence additionally requires fresh a and 0<x\leq v_j. The shared allocation interface must support exactly those x quantity slices. The successor records M(a)=(j,x) and L(a)=\mathsf{Final}. It consumes a once, with \Delta^{\mathrm{eff}}(a)=\mathsf{Pay}(j,x),\qquad R(\mathsf{Pay}(j,x))(u,v)=(u,v-xe_j). The allocation and claim writes commit jointly. The transition preserves every earlier final entry and consumed identity. Its integer-space transformer is total. The admission guard establishes the reachable-state bound.
The reference booking adapter admits a signed recipient-credit root before allocation. Its identity binds book, account, booking epoch, stable event identifier, record kind, asset, and unit. An immutable quantity Q determines slices \{1,\ldots,Q\}. Allocations use disjoint subsets under the same serialized write. The nominated source is fixed by the policy, independently of the reporter. Quantity corrections, alias reconciliation, and actual returns use the shared reconciliation contract in Section 5.2. They preserve the original consumption history. The fixed-root adapter supplies the positive case of that contract.
Derived vesting.
An issued grant schedules each unvested component once its fixed date arrives. The priority is lexicographic in date and component identifier. The due key contains the component and its scheduling generation. The schedule excludes terminal identifiers. Awaiting retains the same identifier and generation. After refusal, a rearm event must identify that refusal and establish current issuer authority and the component’s favorable completion evidence. It increments the generation and preserves all financial projections. The resulting identifier is new. The old refusal remains terminal. Rearm uses the ordinary signed claim interface. Section 3.6 supplies evidence while an earlier component awaits it.
Certificates and validation.
The displayed rules define U independently of any proposed implementation output. A certificate binds the constructor, exact request, premise references, ordered guard results, successor, effect map, evidence changes, and complete writes. Canonical fact order fixes every premise selection. For fixed inputs, there is one normalized certificate. The concrete guard order checks lifecycle and due precedence, request standing, current authority, holder eligibility, effect evidence, quantity, and reporting conditions. Each event omits only guards whose typed applicability predicate is false. This sequence covers its family rule and applicable pack conditions (A1)–(A8). A negative certificate retains the earlier passing witnesses and its first failure or missing-evidence witness. Its private update and effect support are empty. A validation error produces no claim result.
The checker reconstructs the applicable rule from the committed inputs. It checks authority, exact arithmetic, admitted evidence, consumed keys, obligation projection, holder changes, settlement ceiling, and complete write footprint. It compares these values with the proposed certificate before commitment. An optimization can propose a successor. It cannot supply its own acceptance rule.
The producer and checker have different roles. The producer proposes the canonical decision certificate; the checker reconstructs the displayed rule. Correspondence means that checking succeeds exactly for that rule’s normalized derivation, including its ordered negative cases.
Proposition 5.11 (Certificate correspondence).
For either constructor, a certificate passes the checker exactly when it is the normalized derivation of the displayed accepted or ordered negative judgment. The producer terminates on every finite well-typed input that passes ingress.
Proof. The checker branches over the finite event constructors. Each accepted branch reconstructs every rule premise and its exact successor. A successful check therefore supplies that rule’s derivation. Conversely, a normalized derivation satisfies those checks. For a negative branch, the earlier passing guards and first non-pass witness are exactly the negative judgment’s premises. Canonical ordering fixes the certificate in either direction. All parameter sets, evidence sets, and guard sequences are finite. Their typed comparisons, arithmetic, and named evidence checks terminate. The producer therefore reaches one checked decision. ◻
The accounting identity below states that each issued component remains contingent, payable, or paid. The proof checks the event constructors directly, then checks the authority and evidence obligations that arithmetic alone cannot establish.
Theorem 5.12 (Local laws of the concrete constructors).
For every admitted parameter record, both constructors satisfy the ten family laws relative to the specified evidence, authority, and allocation interfaces. For each component of every issued state, u_j+v_j+p_j=q_j.
Proof. Issuance establishes the displayed invariant and accounting equation. Vesting subtracts from contingent quantity exactly what it adds to payable quantity. Payment subtracts from payable quantity exactly what it adds to recorded settlement. The quantity guard preserves nonnegativity. Assignment, pending observation, and rearm preserve these projections. These cases prove invariant preservation and obligation accounting.
Freshness and exact map insertion give one delta per consumed key and monotone consumption. The shared canonicalization contract supplies issuer key discipline. The signed, admitted rule premises establish event authority. Issue and assignment check each new holder. A changed settlement entry uses the admitted ceiling for that exact allocation. Completion supplies no settlement ceiling. The reference scope permits only the fixed engine-audit proposition. An extended scope requires the pack’s explicit permitted-scope derivation. Negative judgments preserve private state and have empty effect support. Certificate correspondence establishes evaluator soundness. These cases establish all ten family laws. ◻
Engine realization.
The reference engine uses typed canonical bytes, immutable digest resolution, Ed25519 request and source authentication, exact integer arithmetic, and a single serialized commit region for claims, nonces, receipts, and allocations. Evidence input uses a separate journal while sharing source admission checks. Receipt construction binds full contingent and payable projections. Historical replay reconstructs decisions from their retained inputs. These operations instantiate the eight engine obligations for the reference execution model, whose state operations terminate without process failure.
A persistent realization also needs durable all-or-none writes and crash recovery. A distributed realization uses the complete dependency and authority protocol in Section 4.2. Its source, provider, legal, and failure assumptions remain named interfaces. The reference engine does not substitute its local mutex for those interfaces. Its tests exercise malformed certificates, competing allocations, stale authority, partial settlement, cure, and replay against the actual constructor code.
6 Historical audit replay and rule change
An auditor may need to explain yesterday’s accepted transfer after today’s rule excludes the same holder. Re-evaluating yesterday’s request under today’s rule answers a different question. Historical replay reconstructs the original decision from the rule and evidence used at its commitment.
The transition interface binds content because a mutable version label has no stable meaning. Historical audit replay is not a new live submission. It runs the evaluator against the content, times, snapshots, facts, and derivations retained for the committed trace. It verifies their authentication and digest links but does not query the current status service or recheck current source heads. A live resubmission must use current heads and can reach a different result.
Assumption 6.1 (Historical authentication).
For replay, each archived snapshot and source-head attestation is verified against the authentication rule and trust material that its committed policy named at the assigned commit time. The archive resolves every bound object by its committed digest. This check replaces the live current-head recheck; it does not assert that the archived head is current now.
Replay first recovers and authenticates the same input content. Deterministic evaluation then gives the same decision and next state. Induction repeats that comparison along the trace. The archive conditions below specify when those complete inputs remain available.
Definition 6.2 (Historical audit replay).
For each committed step, an audit-replay record contains the full request preimage and proposed fact bundle, assigned commit time, retained authenticated snapshot and its current-head attestation at that time, admitted fact set, pre-receipt decision and its certificate, accepted derivation when present, and receipt. Given the initial base and nonce map, \mathsf{Replay} uses the request, proposed facts, bound semantics, assigned time, historical snapshot, and observed state to recompute admission, the decision, and the receipt in trace order. The retained admitted set, decision, certificate, derivation, and receipt are comparison targets, not evaluator inputs. Replay rejects a record whose authentication, digest link, or recomputed value differs. It never queries a current status source.
6.1 Private retention and offline replay
A receipt digest cannot supply the facts needed to recompute a decision. The archive must retain the actual inputs and the ability to authenticate them. Privacy adds a second requirement: the archive must control who can recover those inputs. The following storage rules preserve replay under changes of encryption keys and archive location.
A retention policy \rho names controller, governing source, permitted purposes, recipients, retention intervals, holds, key custodians, and disposal authority. A replay manifest binds the initial base, nonce map, ordered records, terminal root, and their complete transitive input closure. The closure includes semantic objects, proposed facts, historical public keys, authentication rules, source snapshots, and shared allocation evidence. A retained checkpoint defines a replay range. It does not recreate earlier inputs.
Canonical digests remain inside the private archive. Each object is encrypted under a fresh data key with authenticated encryption. The associated data binds its digest and retention policy. Recipient keys wrap the data key for an exact authorized access class. Public references are random opaque handles. Publishing a plaintext hash of a low-entropy private fact can disclose that fact through enumeration, so that hash is not a public privacy mechanism.
An offline grant binds recipient, manifest, purpose, operation, and permitted disconnected use. Its recipient receives the full authorized input closure. Replay verifies historical authentication and canonical digests locally. It performs no current-status request and executes no external action. A policy requiring current authorization instead requires an online access check. Revocation cannot recall plaintext already disclosed under an earlier grant.
Key rotation rewraps each required data key and verifies that every canonical plaintext remains recoverable before retiring the previous managed key path. An archive transfer decrypts, verifies, and imports the complete authorized closure. The receiving archive confirms the same terminal replay result before source disposal. Historical signature verification retains public trust material, rather than old signing keys.
Proposition 6.3 (Replay under archive transfer and key rotation).
If transfer or rotation preserves the complete canonical input closure and its historical authentication, it preserves every replayed decision and terminal root.
Proof. Encryption and key wrapping change storage representations. Successful recovery yields the same canonical evaluator inputs. Deterministic historical replay therefore follows the same induction as Theorem 6.4. ◻
Full replay, selective disclosure, and commitment verification have distinct results. A missing preimage, missing key, denied access, or invalid authentication prevents full replay of any range whose closure needs that object. A tombstone cannot act as the missing evaluator input.
Disposal requires the named authority, expired retention duty, and absence of a hold. Its record identifies exact managed objects, key copies, replicas, and affected ranges. The archive records disposal intent before removal and records each completed removal. Unknown external copies remain outside a claim about completed managed disposal. The remaining commitment can authenticate a later produced record. It cannot establish full replay while the required content remains unavailable.
The executable archive uses AES-256-GCM, per-object data keys, recipient wrapping, private SHA-256 digests, and random public handles. Its offline bundle binds signed source material, the requested manifest, and the exact loaded reference-program digest. The loader fixes the program identity at initialization and rejects later source drift. The tests recover and replay both families with network access disabled. They check access denial, rotation, retention holds, authorized disposal, missing inputs, and ciphertext or manifest alteration. The deployment authority must supply the applicable policy and control its actual recipients, key custody, replicas, and deletion records.
Theorem 6.4 (Deterministic historical audit replay).
Fix a claim instance, initial base and nonce map, and an ordered sequence of audit-replay records with an available complete input closure. Fix the content-addressed family, template pack, rule, and admission-policy semantics named by their active bindings. Under the engine and family laws and historical-authentication assumption, with commit-current admission replaced by historical authentication, \mathsf{Replay} produces the same admitted sets, decisions, receipts, revisions, receipt roots, derived-event dispositions, consumed-effect sets, settlement ledgers, and final private state.
Proof. At step one, immutable digest resolution, the assigned time, and the retained historical snapshot and fact bundle produce the same admitted set. The audit replay verifies the retained current-head attestation rather than querying current heads. Functionality of the family relation and total deterministic pack decision give the same decision. Canonical receipt construction and serial extension give the same next base. Assume equality through step i-1. The private state, revision, receipt root, nonce map, active binding, and observed base at step i are then equal. The request preimage, proposed facts, content objects, time, and historical snapshot produce the same admitted set and pre-receipt decision. The recomputed certificate and, when accepted, derivation equal their retained comparison targets. The next result, receipt, and base are equal. Induction gives equality for the full trace. Exact nonce retries return the already-bound receipt and add no trace step in either replay. ◻
The replay theorem fixes the interpretation as well as the event sequence. A one-threshold holder rule already shows why this matters.
Example 6.5 (Unversioned rule change breaks replay).
Let a transfer rule admit a holder when score k\geq 5. A historical transfer to a holder with score 6 is accepted. The rule later changes to k\geq 7. If the history records only “apply the current holder rule,” replay now refuses the same transfer. One history yields two states.
Pinning the original rule-content digest preserves the old replay. Rebinding the same human-readable version label to changed code does not. Applying the new rule to old history is a different computation. It needs an explicit migration rule and a new state root.
Corollary 6.6 (Replay boundary for retroactive change).
A system that changes a rule affecting committed history has two coherent choices. It can retain the content-addressed old rule for replay, or it can run an explicit migration that binds the new digest and commits a new root. It cannot silently resolve an old label to new semantics while claiming the old state root has unchanged meaning.
6.2 Committed refusals
Refusals matter because they explain why a submitted claim event did not change state. A system that commits only accepted effects cannot distinguish a rejected transfer from a transfer that was never evaluated.
Proposition 6.7 (Refusal audit replay).
Suppose an ordered trace root commits a refusal receipt. Historical audit replay recomputes from its full request preimage and proposed fact bundle, content-addressed semantics, assigned commit time, retained status snapshot and current-head attestation, and observed base. It either reproduces the admitted facts, negative decision certificate, and refusal, or produces a different receipt root unless canonical encoding is ambiguous or an adversary finds a commitment collision.
Proof. The receipt binds the complete request, including nonce and event time, plus the observed base, assigned commit time, status snapshot, admitted-fact digest, reason certificate, and unchanged output state. Historical audit replay uses the retained preimages and yields the same receipt. Any different bound field, decision, or reason changes the receipt commitment. The change propagates to the ordered receipt root unless encoding is non-canonical or a collision is found. ◻
The proposition does not prove that every attempted request entered the log. A censored request leaves no committed receipt. Submission evidence and log-inclusion evidence remain separate requirements.
7 Economic identity through semantic change
The opening obligation still has 40 unpaid units after a change of software. The new program must also retain who may collect them, which 60 units were paid, and which instructions remain unresolved. We call a change of representation that preserves these facts semantic carriage. A duty’s identity comes from the act that created it. A program digest instead identifies the rules that interpret its record.
Consider the split into positions of 60 and 40 units. Assigning all earlier payment to the first leaves that position paid and the second wholly unpaid. A proportional allocation instead places 36 paid units in the first and 24 in the second. Both preserve the total outstanding amount, but they assign different rights to the positions. The migration must specify the portions each position represents and retain their payment history.
7.1 The enduring duty and its representations
An admitted constituting act creates an economic duty identifier d=H(\mathsf{Enc}(\mathit{issuerNamespace}, \mathit{constitutingAct},\mathit{occurrenceIndex})). The issuer’s admission rule identifies the authoritative act and its aliases. It registers an occurrence index once within that act. A changed template, family, price, route, or program cannot register another copy of the same duty. This is an institutional identity premise, strengthened by a unique registry entry. Collision resistance alone does not identify two descriptions of one legal act. A genuinely additional issue receives its own constituting act.
A quantity coordinate lets the record distinguish portions without tying them to a particular family’s position names. Mark each portion with its current rights and whether it is contingent, payable, or discharged. The journal also retains the acts that produced those labels. This retained history is needed when a later return or correction refers to an earlier payment.
Definition 7.1 (Canonical duty journal).
For each quantitative duty d, let X_d=[0,q_d) have rational endpoints and its usual length measure \mu_d. Let \mathcal B_d be the algebra of finite unions of half-open rational intervals in X_d. A canonical journal J records:
- (J1)
-
the constituting identity and current typed rights on a finite partition of X_d, including debtor, beneficiary, asset, unit, due time, performance condition, priority, and governing entitlement;
- (J2)
-
the partition into contingent, payable, and discharged portions, with any other family disposition represented by its own distinct label;
- (J3)
-
admitted changes of rights, external occurrence and allocation history, and separately keyed reversals or reopenings;
- (J4)
-
scheduler decisions indexed by (d,\mathit{eventClass},g), where g is an authorized decision generation;
- (J5)
-
each original command’s canonical duty portions, funding domain, reservation, evidence references, and unresolved outcome.
All these records are part of the committed state. The current rights and labels are derived from its admitted history. The constituting identity is stable through later assignment, amendment, and settlement.
The interval coordinate identifies portions of one duty. It does not identify money in a bank account. A settlement allocation separately identifies the provider occurrence that discharges a portion. For indivisible performance, the same construction uses a finite set with its declared measure. Quantities in different assets, units, or performance classes remain separate.
For a rights tuple t and disposition label \ell, write X_d(t,\ell) for the portions of duty d carrying that tuple and label in the journal. In the 100-unit example, these labeled portions distinguish the paid 60 from the payable 40 before either is assigned to a successor position.
Definition 7.2 (Family representation).
A representation \mathcal R assigns each family position r a finite collection of tagged portions (d,A), with A\in\mathcal B_d. For every registered duty, these portions form an exact partition: \coprod_{r}\mathcal R(r)\big|_d=X_d. A scalar position combines only portions with equal current rights and unit. A structured position retains the tags of its heterogeneous components. For every rights tuple t and disposition \ell, its projection is B_{\mathcal R,J}(r;t,\ell)= \sum_{(d,A)\in\mathcal R(r)} \mu_d\bigl(A\cap X_d(t,\ell)\bigr). A family adapter certifies that its private state decodes to this projection, the journal’s consumed-effect history, and its current decision generations. Its event rules must operate on those same canonical portions.
This adapter is a substantive family proof obligation. Equality of a displayed balance is insufficient. A grant with 100 contingent units and a receivable with 100 payable units have the same total and different obligations. The disposition-indexed projection distinguishes them. It also prevents coalescing two creditors into a scalar balance that conceals who may collect.
An event address is u=H(\mathsf{Enc}(d,\mathit{eventClass},g)). Its execution address additionally records the active binding \beta. An accepted evaluation need not discharge a duty. A refusal never does so by itself. A permitted cure can authorize generation g+1 for the same d, retaining the earlier refusal. Rearming changes an execution opportunity. It cannot create another economic entitlement or clear a payment allocation.
7.2 The carriage transition
Carriage replaces the family representation while keeping the same journal. Its witness must show exact coverage by the old and new representations. It must also establish that the old program loses permission to execute new work when the successor becomes active.
Extend a committed claim base to contain z=(F,\beta,s,\nu,\rho,J,\mathcal R). The family still supplies its private state and laws. The additional journal connects the obligation projections of different families. Its content digest is bound into the state digest and every migration request.
Definition 7.3 (Authorized semantic carriage).
A carriage witness m names the exact observed base, the old and new semantic bindings, both family states and representations, the journal digest, and the complete mutable footprint. The transition (F,\beta,s,\nu,\rho,J,\mathcal R) \xrightarrow{\,m\,} (F',\beta',s',\nu+1,\rho',J,\mathcal R') is admitted when:
- (M1)
-
the current migration authority and receiving-family admission rules authorize this exact source, destination, scope, and transformation;
- (M2)
-
both adapters agree with the same J, both representations cover the same duties exactly, and P_{F'}(s')=1;
- (M3)
-
one serialized commit covers the exact current base, duty registry, consumed effects, pending commands, reservations, and participant discovery;
- (M4)
-
the commit fences new execution under the old binding and activates exactly one successor. The journal and its funding domain remain unchanged.
The new receipt records m and the before and after state digests, with \rho'=H(\mathsf{Enc}(\rho,d(m))). A failed admission preserves the private state and the journal.
The mutable footprint includes the query that discovers affected duties and participants. A changed discovery result, alias map, absence condition, or range invalidates the prepared migration. The joint protocol in Section 4.2 supplies the required commit discipline for remote records. A list of the records found before preparation does not establish that the list remains complete.
The old-binding fence controls new dispatch. A command dispatched before carriage retains its original duty portions and reservation. Its delayed outcome is reconciled under that original binding, even after activation of the new family. The current family sees the resulting change to J. An unresolved old command therefore cannot be forgotten, redispatched as new, or used to release its reservation. Moving execution between stores additionally needs exclusive handover of the canonical registry and the same funding authority. A local receipt cannot supply that remote exclusion by itself.
Theorem 7.4 (Preservation across carriage).
Start from an admitted state with P_F(s_0)=1, a valid canonical journal, and an exact family representation. Suppose the family and engine laws hold within each active binding, each family adapter satisfies Definition 7.2, and every carriage satisfies Definition 7.3. Then any finite trace that interleaves ordinary steps and carriage has these properties:
- (C1)
-
Every carriage preserves the measure of every current rights and disposition class. In particular, it preserves each beneficiary’s contingent, payable, and discharged quantity in each unit.
- (C2)
-
Splitting and coalescing representations cannot create another duty, erase a consumed occurrence, or reactivate a terminal decision generation.
- (C3)
-
A command prepared before carriage retains its canonical allocation and reservation. Its eventual admitted outcome can change the duty once through the ordinary effect rule, independently of the current representation.
- (C4)
-
The private state satisfies its active family invariant, and the committed revision increases strictly at every step.
These conclusions permit authorized economic changes at ordinary steps. They attribute no economic change to carriage itself.
Proof. For one carriage and fixed d,t,\ell, finite additivity on the old partition gives \sum_{r}\mu_d\bigl(\mathcal R(r)|_d\cap X_d(t,\ell)\bigr) =\mu_d\bigl(X_d(t,\ell)\bigr). The new partition gives the same expression. The journal fixes both the measure and the labelled set. Summing over duties of the same type proves (C1). This argument permits different partitions of paid and unpaid portions. It makes no proportionality assumption about their allocation to positions.
The constituting registry, consumed occurrences, and decision-generation history are unchanged in J. Exact coverage assigns every canonical portion to one active representation. Thus neither a changed digest nor a changed position name can produce a new canonical duty or erase its history. The exclusive commit prevents the old and new representations from admitting competing executions. These facts prove (C2).
An unresolved command and its reservation are also unchanged in J. The original command binding identifies the same portions after carriage. The ordinary keyed-effect rule admits its outcome once and records its consumed occurrence in the canonical journal. Every adapter subsequently reads the same changed journal. This proves (C3), including outcomes received after a semantic change. It uses the reservation and exclusion premises and does not infer an external payment from a local migration receipt.
The receiving invariant and incremented revision follow from (M2) and the serialized commit. Ordinary steps preserve their active family invariant by Theorem 5.2. Induction over the interleaved trace proves (C4) and carries (C1)–(C3) across any finite number of migrations. ◻
7.3 Amendment and useful execution
The journal makes a rights change visible as a different operation from relabeling positions. This matters for the transferred or amended remainder in the opening example: the new representation must preserve whichever rights the authorized economic event actually created.
Pure carriage preserves meaning. An assignment changes the beneficiary under its own authority. A novation, maturity extension, conversion, release, or additional issue changes other rights under its own admitted rule. Such an event records an economic transformation and its lineage before or after carriage. A repayment extension cannot be presented as a lossless decoder. An exchange of currencies requires its denomination rule and evidence. The engine can compose these events while retaining the separate justification for each change.
The finite-integer construction instantiates X_d by integer cells. It uses signed carriage witnesses, current authority epochs, exact base comparison, partition checks, and one serialized journal. It reserves payable duty portions. The funding-domain field identifies the external cash authority and does not implement its cash ledger. A 100-unit coupon paid before a template change remains discharged. A 60-unit payment can pass into a paid 60-unit position and an unpaid 40-unit position. The remaining payment can then settle, and both positions can coalesce. A command prepared before the change can settle afterward against its retained duty reservation.
The same construction reads the concrete receivable and milestone-grant states of Section 5.4. It retains their source state, trace, and occurrence allocations. The receivable continues from 60 paid units to full payment. The grant retains its contingent 100 units across a semantic change. This read adapter does not deactivate the source engine. A deployed handover must realize (M3)–(M4), the family adapters, authenticated provider semantics, and the institution’s actual migration authority.
7.4 Durable execution through a changed representation
The carriage theorem specifies the required agreement between representations. To execute payments after the change, a concrete service must enforce that agreement on every write. The resolved-duty construction does this by keeping one original entitlement ledger. A successor position acts as a payment alias, an address for selected portions of an existing duty.
A representation becomes operational when a payment addressed to it changes the original duty. This requires a common write authority for the entitlement ledger and the portion journal. Comparing two independently maintained balances does not supply that authority.
Here the construction specializes to a finite cohort whose duties have already resolved into payable amounts. Adoption places that original ledger and its portion journal under one write authority. The payable and discharged labels used here are the corresponding part of the more general journal; future unresolved obligations retain their original lifecycle.
Consider a finite cohort of resolved duties in an executable payoff ledger. Each duty retains its original program, constituting identity, directed parties, governing terms, and maturity. Fix a positive rational quantum for each asset. The canonical portion journal measures each duty in those quanta. Its labels record current payable and discharged portions. Historical allocations and their later authorized reopenings remain separate records.
A successor program executes on its own temporary ledger with the declared inputs. Each resolved output names portions of the original cohort. The admission checks every emitted duty, the exact partition, and the matching economic fields. It retains the successor checkpoint and its evidence. The resulting output names are payment aliases for original duties. The temporary duties remain outside the active entitlement ledger.
An alias payment selects payable portions owned by that alias. The service then invokes the original ledger’s allocation operation for their total. One transaction commits the allocation, its physical support, and the selected portions. Ordinary payments and authorized reopenings use the same write mediator. An outstanding command retains its original target and recipient. Its eventual outcome updates the original duty through this common authority.
Proposition 7.5 (Durable resolved-duty correspondence).
Suppose one authenticated store owns the original ledger and its canonical portion journal. Admission binds the exact current state and a complete successor partition. Every successful mutation uses the common write mediator. Then, after each committed mutation, for every adopted duty d, \operatorname{Outstanding}_L(d) =\delta_d\,\#\{i:\operatorname{Label}_J(d,i)=\mathsf{Payable}\}, where \delta_d is its asset quantum. Alias installation preserves the active duties, allocations, and reopenings. A payment consumes its selected payable portions once. A committed retry preserves the same allocation and outcome.
Proof. At adoption, retain the original ledger. For each historical allocation, subtract its authorized restorations to obtain its current discharge. Assign distinct cells to these remaining discharges. The other cells are payable, so the displayed equality holds. Retain historical restoration coordinates separately, including cells discharged again by later payments.
Installation changes only the active representation and its retained evidence. The exact partition preserves every payable quantity. Each alias allocation chooses payable cells within that alias and calls the ledger on the original duty for exactly their measured quantity. Both sides of the equality decrease by that quantity in the same commit. An authorized reopening restores cells of its particular allocation and increases both sides equally. Other ledger operations preserve the adopted duties’ outstanding amounts or pass through these same cases. Physical allocation history retains its consumed slices. A later reopening changes the duty label while preserving that history.
The transaction commits the ledger, portion changes, and command outcome together. Before commit, recovery restores the preceding state. After commit, the recorded command identity and full request recover the same outcome. A retry therefore adds no allocation. Induction over committed mutations proves the claim. ◻
The reference service adopts the existing SQLite store under one write transaction. Its successor schema retains original program checkpoints, ordered ledger records, command outcomes, and authentication data. The old client checks the schema within its own write transaction. It therefore rejects a write after adoption. A write committed first changes the base and invalidates the prepared adoption. The successor service continues original requests under their original program identities.
An executable coupon illustrates the distinction. Of a 30-unit duty, 20 units are discharged. A successor program produces a paid 20-unit alias and a payable 10-unit alias. Payments of five units from remaining physical support and five units from another credit complete the original duty. A second program coalesces the aliases into one discharged position. The original ledger contains the same duty throughout. Another unfinished coupon request then resolves under its original program.
The service executes this trace through the payoff evaluator and entitlement ledger. It retains a partially settled operation’s target, source digest, command, and reservation across adoption. Recovery retains the same committed outcome when the payment response is lost. The interval construction below represents large quantities while preserving these allocation histories.
This construction covers resolved duties with fungible canonical portions. It leaves correction duties as distinct directed obligations. A later original-program resolution retains its ordinary lifecycle outside the adopted cohort. Transfer of unresolved computations requires a further constructor. Section 7.6 supplies one for programs with the same instructions and rational position partitions. Arbitrary future-input equivalence remains outside these constructions. Actual migration authority, provider facts, and exclusion between independent stores remain institutional premises. The local transaction establishes the common write authority within the declared store.
7.5 Exact interval representation
The resolved-duty construction assigns a coordinate to each smallest permitted quantity, or quantum. Enumerating those coordinates makes storage grow with monetary quantity. Intervals can store consecutive portions together, provided they retain the selections and history used by later payments and returns.
Consider four units paid by one allocation. Two units are returned and become payable again under the governing rule. A replacement payment discharges those two units. A further return against the first allocation must concern its remaining discharge; it cannot reopen the replacement payment’s units.
Number the original units from zero through three and call the allocations a and b. A return against a reopens cells 0 and 1, which b then discharges again. A further return of one cell against a must reopen cell 2. The current payable balance alone cannot determine that choice.
An alias row is the ordered list of portions addressed by one successor position. A triple names a duty and the beginning and excluded endpoint of one consecutive portion. Thus a partial payment can depend on the row’s order even when two rows cover the same portions.
Order also matters before a payment. When all four cells are payable, the alias row ((d,2,4),(d,0,2)) first selects cell (d,2). Sorting its intervals first selects (d,0) instead. The rows have equal support and quantity, but different partial-payment behavior. We therefore distinguish ordered sequences from membership sets throughout the construction.
Definition 7.6 (Ordered runs and supports).
Fix a finite resolved cohort with a positive rational asset quantum \delta_d and a positive integer quantity q_d for each duty d. Its canonical cells are (d,i), where 0\leq i<q_d. A run (d,a,b) denotes the sequence (d,a),\ldots,(d,b-1). For an ordered run list R, let E(R) concatenate these sequences. Write S(R) for its underlying support and |R|=\sum_{(d,a,b)\in R}(b-a). Every admitted list has integer endpoints and distinct cells within the declared duty domains.
A membership set has sorted disjoint maximal runs within each duty. An ordered list retains the order of its entries. Only consecutive entries (d,a,b),(d,b,c) can coalesce without changing their expansion. Duplicate portions are rejected before normalization can conceal their overlap. Equal indices belonging to different duties remain distinct coordinates.
The following operations act on ordered lists. Intersection and difference first express the mask as a sorted disjoint union. Within each source run, they split at the mask endpoints and retain the appropriate successive subintervals. They visit source runs in their original order. The prefix operation visits runs in order and consumes their lengths until the requested count is reached. Its last run is shortened when necessary.
A membership mask specifies which cells qualify, without prescribing their order. Filtering an alias by that mask must preserve the alias’s order. The lemma states this fact and then identifies a partial payment with a prefix of the resulting sequence. Square brackets on the right denote order-preserving sequence selection or a prefix.
Lemma 7.7 (Ordered selection).
Let R be an admitted run list, M a membership mask, and 0\leq k\leq |R|. The intersection, difference, and prefix operations satisfy \begin{align*} E(\operatorname{intersect}(R,M)) &= [x\in E(R):x\in S(M)],\\ E(\operatorname{subtract}(R,M)) &= [x\in E(R):x\notin S(M)],\\ E(\operatorname{take}(R,k))&=E(R)[0:k]. \end{align*} In particular, the first k payable cells of an alias R with payable support P are E\bigl(\operatorname{take}(\operatorname{intersect}(R,P),k)\bigr), \qquad 0\leq k\leq |S(R)\cap P|.
Proof. The mask endpoints partition each source run into elementary intervals. Membership is constant on each such interval. Intersection retains precisely the intervals inside the mask, and difference retains their complement. Both retain increasing order within the source run and the original order between source runs. This proves the first two identities.
The prefix algorithm consumes exactly k cells. Every consumed cell precedes every unconsumed cell, which proves the third identity. Coalescing adjacent consecutive runs preserves their expanded sequence. Applying the first and third identities gives the final formula. ◻
For each duty, a label partition covers [0,q_d) with disjoint intervals marked \mathsf{Payable} or \mathsf{Discharged}. Relabeling splits at the selected endpoints and changes precisely those portions. Adjacent intervals with equal labels then coalesce. An alias installation retains the complete, disjoint coverage of every duty and each row’s order. It checks the parties, governing terms, quantum, maturity, context, and all other required economic fields. The quantity of each label in an alias is the length of their intersection, multiplied by the quantum. Finite additivity therefore preserves the aggregate quantity of each duty and label across all aliases. The exact economic bindings preserve the corresponding rights classes.
A duty label answers whether a portion is currently paid. A restoration record answers which earlier allocation has had that portion returned. The four-unit example requires both answers: a portion restored from the first allocation can now be discharged by the replacement allocation.
For allocation a, retain its ordered original sequence A_a and its own restored support R_a. Its current discharge support is L_a=S(A_a)\setminus R_a. Historical supports S(A_a) can overlap after a return and replacement payment. The supports L_a must remain pairwise disjoint. Their disjoint union is the discharged support, whose complement is payable. In addition, R_a\subseteq S(A_a) for each allocation.
For the four-cell example, the original sequence of a is (0,1,2,3). After its first return, R_a=\{0,1\} and L_a=\{2,3\}. The replacement allocation b has original sequence (0,1) and empty restored support. Its current discharge is L_b=\{0,1\}. A further return against a selects the first cell of its remaining sequence, namely cell 2, while retaining the discharge recorded by b.
Alias rows determine selection order, and labels determine current payable status. Allocation sequences and restoration records determine which earlier payment a return can reopen. The separate physical support below records which portions of an external credit have already been consumed.
Adopting a ledger assigns portion coordinates where it previously retained only quantities and allocation history. Converting an existing cell state instead preserves its coordinates and order. The refinement proof treats these two entry cases separately.
The adoption rule constructs these supports from the existing ledger in its retained allocation order. For an allocation with total t and restored quantity r, choose t-r cells from the remaining free support. Then choose r historical cells from the duty domain outside that allocation’s selected active support. The allocation stores the active sequence followed by this restored sequence. Only its active support is removed from the free support. Prefixes and differences perform both selections without enumerating their cells. Thus active discharges remain distinct while historical restoration coordinates can be reused by later allocations.
A new payment selects a prefix from the alias’s payable support. It creates an allocation with that ordered sequence and empty restored support. A return against allocation a instead selects T=\operatorname{take}(\operatorname{subtract}(A_a,R_a),k). It adds S(T) to R_a and marks those cells payable. The original ledger’s return evidence and governing-rule checks authorize this change. An interval membership test supplies no such authority.
Physical credit has a separate coordinate (o,j), with 1\leq j\leq Q_o. Here o identifies the canonical external occurrence admitted by the original ledger. A legal origin, beneficiary, or payment alias does not change that occurrence identity. Consumed physical supports remain disjoint within each occurrence. Their union remains consumed after a duty reopens. A replacement payment therefore requires other unconsumed physical support and its admitted evidence. Explicit physical selections retain their supplied order and pass extent and overlap checks.
For an alias spanning several duties, selected cells are grouped by duty in first-occurrence order. Each group retains its internal selection order. The physical sequence is then split by the successive group quantities. This rule fixes the allocation pairing when an alias revisits a duty later in its ordered row.
The comparison cell construction uses the stated transition rules on arbitrary finite cell sets. It imposes no additional bound on the number of cells. It supplies a mathematical interpretation, while the interval representation executes without enumerating that interpretation.
Refinement means that expanding the stored intervals gives the same execution as representing every cell individually. The proof uses ordered selection for payments and returns, and disjoint supports for current discharges. Physical payment portions have a separate consumption history, which reopening a duty must preserve.
Theorem 7.8 (Interval refinement of resolved-duty execution).
Use the finite resolved cohort of Definition 7.6 and the exact economic and admission conditions of Section 7.4. Suppose one authenticated serialized store owns the original ledger and the portion journal. Every mutation passes through the common write mediator, and the ledger’s admission rules remain unchanged. Then the interval construction simulates the cell construction after every committed mutation. Expansion gives the same labels, ordered allocation selections, restored supports, and consumed physical supports. At corresponding states, the constructions have equal original duties, program checkpoints, economic outcomes, and retained prior receipts. Each payment or authorized return has the same original-ledger effect. In particular, Proposition 7.5 holds for every finite quantity represented by the intervals.
Proof. Adoption chooses each active discharge from the remaining free support. These supports are disjoint, and their union receives the discharged label. The allocation’s restored sequence lies outside its own active support. Consequently S(A_a)\setminus R_a is exactly its selected active discharge. The ledger’s outstanding amount equals \delta_d times the remaining payable cell count. Lemma 7.7 gives the same selections as the cell construction.
An existing cell state converts by retaining its ordered runs, maximal label intervals, and physical membership sets. Each expansion equals its former sequence or support. The conversion preserves the original ledger records, program checkpoints, and previously committed outcome objects. Installation changes only the aliases and their retained evidence. Exact coverage and unchanged economic predicates preserve each duty’s aggregate rights and label quantities across the aliases.
A payment selects only payable portions. Its new active support is therefore disjoint from every previous active discharge. The selected cells become discharged, and the original ledger receives their exact measured quantity. Ordered filtering, grouping, and physical partitioning reproduce the cell allocation’s arguments. An authorized return selects only the named allocation’s unrestored support. It removes that support from exactly one current discharge and adds the same quantity to the original outstanding duty. Other allocations retain their own restoration history. These operations preserve the stated partition and Proposition 7.5’s equality.
Physical synchronization retains the original occurrence identity and consumed support. Reopening changes duty labels without releasing that support. Other mediated mutations retain the original ledger and program behavior or use these same payment and return cases. A previously committed request returns its retained outcome before any new base check.
One transaction commits the ledger, interval state, and keyed outcome together. A failure before commitment restores the preceding related state. A completed commitment stores the related successor and its recoverable outcome. Induction over committed mutations proves the simulation. ◻
The decoding theorem compares economic state and ordered selections. A new receipt can encode a selected sequence as intervals. Its serialized bytes then differ from a receipt that lists the individual cells. Converting the store preserves previously committed receipt objects and their request bindings exactly. Different orders of physical selection remain different requests even when their supports agree. Old writers check the active schema within their own transaction, so that the adopted store retains one write authority.
Quantity and computational cost.
Consider an ordered intersection, difference, or prefix operation, with normalized membership masks. Let n count its input and output runs. Let b bound the bit length of its integer endpoints and accumulated quantities. The direct nested interval algorithms use O(n^2) integer comparisons and additions in the worst case. Elementary integer arithmetic therefore gives an O(n^2b) bit bound for these primitives. Exact quantum conversion has the additional cost of rational arithmetic. Authentication and durable serialization also depend on the size of retained records and history.
One uninterrupted amount needs one run even when its quantity is large. Independent allocations and returns can create many intervals, and their histories must remain distinguishable. Work thus depends on fragmentation, retained history, and endpoint bit length. It does not require a loop over every represented quantum. The construction removes the enumeration bound while preserving the represented economic distinctions. A concrete store still enforces its record and snapshot limits. The construction establishes neither a bound on institutional throughput nor a financial valuation.
7.6 Position partitions across future observations
A coupon can depend on observations that arrive after its positions change. Its present balance alone does not carry that unfinished computation. The following constructor retains the program’s instructions and changes its position partition. It preserves historical dependencies and later requests, including requests beyond those pending at admission.
The source program continues against the current entitlement ledger. A second evaluator runs the successor program against a private ledger. Both receive the same admitted requests and observation refinements. Either may finish first.
A source duty can receive payment while the successor remains suspended. Once both evaluators resolve the instance, their paired results give a correspondence with original duties. The conditions below determine when it can also become a payment alias.
Definition 7.9 (Structural position certificate).
Let P and Q be authentic compiled programs admitted by the same typed payoff evaluator. Their contexts, oracle contracts, claim names and order, schedules, declarations, instruction graphs, output types, and deadlines agree. Their bindings of execution authority, scope, and purpose also agree. Within each claim, only positions may differ. The permitted instructions are constants, oracle observations, historical references, window reductions, linear combinations, multiplication, indicators, caps, floors, conditionals, and exact rescaling. Currency conversion, resolved-rate references, and external conversion witnesses are excluded.
A position family is its complete tuple of issuance, lot, and leg identifiers. Each family occurs once within a program. The source and successor families are disjoint. A position’s economic signature retains every field except its leg identifier and quantity. It thus retains issuance, lot, parties, asset, terms, evidence, and governing and payment contexts.
Within each claim and signature, write the source quantities as q_1,\ldots,q_m>0 and the successor quantities as r_1,\ldots,r_n>0. They are exact rationals with equal totals. Place each list consecutively on the same interval, giving partitions (I_i) and (J_j). For every pair, define a_{ij}=|I_i\cap J_j|. The certificate records each positive entry together with its claim and both complete position families. Verification reconstructs these ordered intersections and checks exact certificate equality. It also binds both compiled programs and their evaluator semantics.
For each source and successor position, interval additivity gives \sum_j a_{ij}=q_i,\qquad \sum_i a_{ij}=r_j. \tag{1} Thus several source positions can regroup into several successor positions. The certificate preserves their individual original identities through the intersections. Within this construction, equality of compiled expressions supplies the required agreement of numerical meanings. Authenticity of the admitted program objects remains a premise. A digest check alone does not validate hostile changes to their internal structure.
Retaining unfinished computation.
An auxiliary claim records a reusable historical value without a payment position. Admission reads the source checkpoint and current ledger under the write authority. It checks every occupied source family against its exact resolution witness. It checks each occupied reserved auxiliary name against its source identity and amount. The canonical ledger must contain no successor family.
These checks exclude a foreign binding at a future requested height. The authority reserves source families for source execution and keeps successor families outside the canonical ledger. Unrelated lawful ledger operations remain available.
Historical dependencies refer to strictly earlier heights under the evaluator’s schedule contract. Transport therefore visits retained resolutions in increasing height order. It checks each retained resolution identity and historical reference. It replaces each historical identity by the corresponding successor identity already constructed, then computes the successor resolution identity. The source records retain their original identities and evidence.
The successor’s private ledger receives its own bindings and auxiliary records through the ordinary entitlement-ledger operations. Cached values use the same checked historical-identity map. Requests, pending tasks, cache invalidation counters, and instruction indices carry across unchanged. Transport also copies the ledger’s mutable map of auxiliary memory values. Replaying auxiliary records in height order need not reproduce a map created in another execution order. Historical evaluation continues to read immutable resolutions.
The successor inherits consumed source work as budget debt. Its input charge adjusts for the difference in encoded program sizes. This transfer is not evidence of a second execution of earlier instructions. Admission and execution remain subject to finite checkpoint, materialization, and evaluator limits. Transport and serialization costs lie outside the evaluator’s arithmetic counters.
Two prefixes are compatible when they belong to the same admitted observation-refinement history. Missing inputs may arrive in that history. Replacements contract rational intervals with new evidence, and exact observations remain immutable.
Lemma 7.10 (Agreement after compatible refinements).
For programs satisfying Definition 7.9, transport preserves retained numerical values and types. Suppose both evaluators later resolve the same instance from compatible prefixes of admitted observations. Their exact number, type, and numerical or Boolean sort agree. Their evidence identifiers need not agree.
Proof. At transport, induction on height validates each historical dependency before its use. Equal instruction graphs and declarations give the same numerical meaning to every transported value and instruction position.
For later execution, order interval information by reverse inclusion. An admitted replacement contracts its interval with new evidence. An exact observation is immutable. For Boolean information, unresolved information lies below either certified truth value.
Interval addition, multiplication, scaling, minimum, and maximum preserve enclosure under contraction. A certified comparison keeps its truth value. An unresolved conditional selects no branch. Once its condition is certified, compatible refinements preserve the selected branch. Window reductions retain these properties, and historical reads use immutable exact values.
Induction over instructions, then over dependency heights, proves refinement stability. Any later compatible enclosure of a singleton lies inside that singleton. Two compatible prefixes that resolve the instance therefore give the same exact value and type. Each evaluator retains its own observations and recursive resolution identities, so this argument requires no equality of evidence bytes. ◻
Theorem 7.11 (Structural continuation with one economic ledger).
Assume Definition 7.9, authentic admitted source state, and the transport and reservation checks above. One exclusive write authority owns both continuations and the current canonical ledger. Both evaluators receive the same admitted requests and compatible refinements. Each command identity binds its complete request and one immutable outcome, checked before new execution. Then every finite command trace within the declared resource limits has these properties:
- (S1)
-
Whenever both evaluators resolve an instance, their canonical values agree. For a numerical output, let v be its exact number multiplied by its declared positive scale. Intersection (i,j) then denotes the nominal amount a_{ij}v under the original source duty identity.
- (S2)
-
Within the certified position families, only source execution creates canonical economic bindings. Successor execution and publication of a correspondence have no canonical economic effect. They are stuttering steps under that economic projection.
- (S3)
-
Payments, corrections, returns, reopenings, reservations, and cash records retain their original identities and current ledger effects.
- (S4)
-
An exact committed retry returns its retained outcome without further work or state change. Reuse of its command identity with changed contents is refused.
The projection in (S2) omits revisions, authority handles, counters, and receipts. It asserts no equality of run reports or work counters.
Proof. Transport establishes related computations and preserves the canonical ledger. A request or observation admitted to both evaluators has no direct economic effect. A successor run changes only its private computation and ledger. A source run uses the original program against the current canonical ledger. These cases establish (S2).
When both computations resolve, Lemma 7.10 gives the same value. Equation 1 partitions each original nominal amount. Negative values use the unchanged rule that reverses debtor and creditor. Zero retains its resolution binding and creates no payable duty. Publishing the paired records creates no second canonical duty. This proves (S1).
A lawful ledger command uses the current ledger through its ordinary admitted rule. The authority never restores an economic snapshot from admission. A payment made before correspondence completion therefore remains attached to the original duty. It changes the outstanding balance, while the correspondence still partitions the original nominal face.
A target-amount correction creates its ordinary distinct directed correction duty. It does not replace that nominal partition. Family and auxiliary reservations prevent conflicting bindings while allowing unrelated operations. These facts prove (S3).
Each command identity binds the complete request and its retained outcome. Replay checks that binding before applying current-state conditions. An exact replay performs no new action, while changed contents fail. These cases prove (S4). Induction over the command trace establishes all four properties, including traces where either evaluator remains suspended while the other continues. ◻
The theorem applies uniformly to finite prefixes of an infinite schedule. Completion requires finitely many requested instances and historical dependencies. It also requires sufficient admissible resources, sufficient future observations, and fair execution of the pending work. The finite reference evaluator does not promise unlimited execution. A wider position partition can require more work at commitment.
Durable ownership and declared failures.
A finite reference construction realizes this continuation in the authenticated SQLite carriage store of Section 7.4. Admission acquires the write lock before checking the signed request against the exact current predecessor. The signature identifies the policy, source instance and checkpoint, and structural certificate. The new owner retains existing instance records, ledger history, and completed command receipts. An unfinished transfer owned by another continuation must complete through that owner before this admission can proceed. Older writers check the active schema inside their write transaction.
A computation commitment covers every source computation field. It excludes only handle revision and appended ledger history, which lawful ledger operations can change. Every later write checks this commitment, source and auxiliary witnesses, and the absence of canonical successor bindings. The inherited carriage checks continue to enforce the interval invariants. One transaction stores both checkpoints, the current economic ledger, private successor state, paired resolutions, reservations, and the exact command outcome. A message authentication code covers the complete encoded snapshot.
A budget suspension retains the actual partial computation and charged work. A declared execution error instead restores the preceding computation and economic state while retaining attempted resource charges and execution counts. It also retains the attempted resource envelope and invalidates a stale run cache when those charges or limits change. The same transaction records the error type and arguments as a rejected outcome. An exact failed retry returns that outcome without another charge. This atomic failure contract differs from an unwrapped evaluator that can retain partial economic effects before a later error.
Before commit, process loss restores the preceding transaction state, including its resource debt. Work performed only in that lost process is not recovered. After commit, restart recovers the retained checkpoints, debt, and receipt. The caller supplies the exact source and successor programs on reopening, and the store checks their bound identities and certificate. These transaction cases extend Theorem 7.11 by induction over committed writes and recoveries. They assume the configured runtime, authentication keys, signing policy, and durable storage semantics.
Corollary 7.12 (Attaching a future resolved partition).
Under Theorem 7.11, suppose a matched resolution with at least one payment position has positive canonical value v. Its source duties have no existing carriage roots. Retain the adopted positive asset quantum \delta and unit. Suppose every original due amount is integral in that quantum and \frac{a_{ij}v}{\delta}\in\mathbb Z_{>0} \qquad\text{for every positive intersection }a_{ij}. Suppose the current history passes the exact interval initialization checks of Section 7.5. Each historical allocation and its aggregate authorized reopening must have integral quantum coordinates. Attachment checks the exact current base and reconstructs this history. Then the successor portions can become ordered interval aliases for the original duties. Proposition 7.5 and Theorem 7.8 hold for the newly carried cohort.
Proof. For each original duty, order its intersections and assign consecutive intervals of lengths a_{ij}v/\delta. Equation 1 gives a disjoint complete partition of that duty’s integral quantity. Initialize payable and discharged labels from its current original allocations and reopenings. Attachment adds only aliases. It creates no successor economic binding. The existing interval correspondence and common write mediator then give the two cited results, including later payments and authorized returns. ◻
For example, a later observation can resolve a three-unit coupon whose successor quantities are one third and two thirds. With a one-unit quantum, its one-unit and two-unit portions admit attachment. The same quantities applied to a one-unit coupon retain exact rational correspondence but fail this attachment check. Integral face amounts alone also do not ensure that prior allocation and reopening quantities admit integral coordinates. A uniform certificate that every future observation will satisfy a chosen monetary lattice remains a separate problem.
The rational correspondence also covers signed and zero outcomes. This attachment operation covers positive integral outcomes from previously uncarried duties. Existing correction duties retain their distinct lifecycle. General expression equivalence, conversion witnesses, and safe reuse of position families require further certificates.
Actual migration authority and provider facts remain institutional premises. The storage result assumes one monotonic database authority. Whole-database rollback requires an external anchor. The construction supplies neither exclusion across independent databases nor atomicity with an external payment system.
8 Further claim families
These sketches identify the private state and obligation differences that the interface must preserve. They are not admitted instances: each would still need a registered pack, a concrete transition relation, and proofs of the family laws.
8.1 Fixed income
A fixed-income family stores dated principal and coupon obligations. Events include issue, transfer, record-date capture, coupon determination, payment, amendment, default, and redemption. The obligation projection is a dated multiset. A payment delta removes only the obligation supported by final settlement evidence. An amendment needs the authority named by the instrument. A transfer needs the holder rule that applies at the event time.
ACTUS already gives a detailed algorithmic account of contract events and cash flows [1]. The present interface can consume such an event model as a family. It does not replace it.
8.2 Parametric insurance
An insurance family stores coverage, trigger source, limit, deductible, premium status, notice conditions, assessment state, and payment state. A trigger attestation can move the claim into assessment. It cannot itself establish coverage, satisfy notice, or make payment final. The family invariant caps aggregate final payments at the covered limit after deductible and permitted reinstatement rules.
The common engine sees only that the family transition and laws passed. It does not reduce an insurance claim to a binary token.
8.3 Trade receivables
A receivable family stores seller, buyer, invoice, acceptance status, defences, assignment status, maturity, and collections. Assignment changes the holder of a right. It does not remove a buyer defence unless applicable law and the instrument permit that effect. Collection reduces the outstanding obligation only on the family’s accepted finality evidence.
This family shows why a generic holder field is not enough. Family-private state records whom the engine treats as the current holder and which defences and assignment limits remain represented.
8.4 Options
An option family stores writer, holder, underlying, strike, exercise window, style, payoff rule, collateral rule, and settlement route. Exercise is a family event. The payoff language can describe the amount [28]. The lifecycle interface decides whether exercise was authorized and timely, whether settlement evidence is strong enough, and what obligation remains.
The same request and receipt structure can surround each of these state models. Their transition relations must still establish the authority, accounting, consumption, and evidence laws for those particular obligations.
9 What simpler records omit
The constructions retain several kinds of record because each answers a different question. Omitting one gives the following specific failures:
- Erased claim meaning.
-
One balance can conserve quantity while erasing seniority, holder restrictions, conditions, amendment rights, and recovery paths. One claim-wide settlement flag also cannot represent a final 60-unit effect beside a pending 40-unit effect. The lost distinctions cannot be recovered after reduction.
- Duplicated family engines.
-
One root engine per product preserves meaning but duplicates finality, refusal, evidence, and receipt rules. The copies can disagree. The parametric interface keeps family meaning separate while sharing only common obligations.
- Evidence and finality conflation.
-
An engine receipt proves an engine decision, not external finality. PFMI Principle 8 requires an FMI’s rules to define final settlement and the point of irrevocability [29]. Articles 3 and 5 of Directive 98/26/EC likewise make enforceability and irrevocability depend on the applicable system rules [30]. Two messages can also describe one provider occurrence. Route-specific evidence orders and one canonical consumed key preserve these distinctions.
- Serialization and mutable context.
-
An advisory state digest permits competing successors and ABA reuse, in which a changed state returns to an earlier value. An actor-controlled time can conceal expired authority. Hidden scheduled events can mutate state without a trace step. A mutable rule label turns audit replay into re-adjudication. Non-repeating bases, engine-assigned commit time, separate derived-event requests, and active content digests address these failures.
10 Relationship to prior work
The interface uses established work on financial semantics, authorization, and serialized execution. Its trace theorem is conditional and elementary. The comparison below identifies which inputs and distinctions the joint interface requires around those existing mechanisms.
Financial contract languages.
Peyton Jones, Eber, and Seward define compositional financial contracts and a denotational valuation semantics [28]. Findel gives a declarative derivative language for blockchain execution [8]. Marlowe gives executable financial agreements with a small semantics [26]. These systems establish that product terms and payoffs can be expressed compositionally. The present work starts after expression. It models authority, lifecycle state, non-acceptance, external finality, and reliance around the family semantics.
LexiFi’s Contract Description Language also gives self-contained, compositional descriptions of financial contracts. It keeps the mapping from abstract to actual parties outside the language and treats extraordinary events outside nominal contract behavior [19]. Those boundaries agree with the separation here between family semantics, external facts, and the engine envelope.
Algorithmic standards.
ACTUS Technical Specification v1.1 represents contract types through state variables, scheduled events, state-transition functions, and payoff functions [1]. The ISDA Common Domain Model Design Definition v1.0 standardizes product and lifecycle concepts and describes parametric processing [10]. Both are close prior art. This paper does not claim a wider product taxonomy.
FpML is ISDA’s XML business-information exchange standard for derivatives. Its Version 5.13 recommendation covers pre-trade, trade, post-trade, and lifecycle processes [9]. It can carry messages around a family implementation; it does not supply the serialized safety contract proved here.
The FINOS Common Domain Model now maintains a modular representation of financial products, trades, and lifecycle events, with mappings and executable logic [11]. Its event vocabulary can supply family semantics. The joint contract studied here instead asks which external facts, occurrence identities, commit-time checks, and reliance limits a serialized evaluator must bind around those semantics.
Concurrency and idempotence.
Optimistic concurrency control, linearizability, and idempotent message handling are established techniques [16, 13, 12]. The revision, nonce, and occurrence-key mechanisms in this paper are applications of those techniques. The contribution is their typed coupling to W_F, C_F, effect-indexed evidence, and the family laws. It is not the invention of nonces, optimistic validation, or linearizable storage.
Payment-message identity and status.
ISO 20022’s pacs.002 message reports a positive, negative, or pending status for an earlier payment instruction [14]. SWIFT’s 2026 cover-payment guideline requires the pacs.009 COV to carry the same Universal End-to-End Transaction Reference as its underlying pacs.008; the two messages describe the same transaction [32]. Both support the separation used here: a message identifier identifies an assertion, while a stable external reference can identify the occurrence described by several assertions. Neither standard by itself proves that the occurrence is final under the relevant route.
Ledger and workflow platforms.
Corda combines agreement states, contract code, required signatures, time windows, external data, consuming transitions, and a uniqueness service [2]. Daml combines template-private state, consuming choices, authorization, delegation, and compositional multi-party workflows [3]. Hyperledger Fabric validates read-set key versions against committed state before applying writes [4]. These systems already establish important parts of authorization, single consumption, and stale-state rejection. The present paper does not claim those mechanisms separately. It states one joint interface between family-private claim semantics and a serialized engine, including commit-time status, effect-key support, effect-indexed settlement evidence, and bounded reliance metadata.
Credential status.
The W3C Bitstring Status List distinguishes status purposes such as revocation and suspension and requires verifiers to authenticate status-list credentials [33]. It confirms that issuer status is a separate, time-sensitive input. This paper uses a more general authenticated snapshot abstraction because authority, provider, and rule-policy status need not be verifiable credentials.
Smart legal contracts.
Clack, Bakshi, and Braine distinguish operational code from legal agreement and enforcement [6]. Their separation is essential here. The external-fact interface prevents an operational engine from presenting its own result as legal authority.
Typestate and proof-carrying execution.
Typestate systems make permitted operations depend on current state [31]. Proof-carrying code lets a consumer check a producer-supplied safety argument against a fixed policy [27]. The transition envelope applies the same broad discipline to a claim lifecycle. The family supplies the policy and state relation. The evaluator checks the relation before mutation.
Event logs and state-machine replication.
Deterministic state-machine replication requires replicas to apply the same ordered operations under the same semantics [17]. The audit replay theorem here is elementary and conditional. Its purpose is to enumerate the content, time, status, nonce, and evidence inputs that must be fixed when the state machine consumes institutionally sourced facts.
11 Limits and open problems
11.1 Legal adequacy
The family authority predicate can be wrong. A rule translation can omit a statute or misread a contract. A status service can omit a revocation or authenticate the wrong issuer. Formal preservation under wrong content or a false snapshot is still wrong. Legal adequacy and status-source governance need independent validation.
11.2 Realizing the engine assumptions
Canonical encoding, registry immutability, trusted time, and truthful source evidence remain explicit premises. Section 4.2 constructs a prepared transaction for current recorded heads and joint writes. Section 5.2 supplies its occurrence-attribution and funding contracts. A concrete implementation must establish refinement, including recovery, expiry, dependency completeness, and concurrent reservation. Its evidence must include linearizability checks, failure-injection tests, and exact cryptographic and wire specifications. The source-head result does not establish legal authority at a later external use stage.
11.3 Insolvency and priority
Outstanding-obligation accounting does not determine priority in insolvency. It also does not create perfection, property rights, set-off, or close-out netting. A family can record those asserted states only through accepted legal evidence. Their effect remains jurisdiction specific. For example, sections 238–241 of the Insolvency Act 1986 provide statutory consequences for transactions at an undervalue and preferences, including orders that restore the position [15]. An earlier engine receipt cannot displace that power.
11.4 Privacy
Section 6.1 separates private full replay from public references and selective disclosure. Encryption protects stored content under its key and access assumptions. An authorized recipient can retain disclosed plaintext. A deployment must establish its actual access grants, retention duties, and managed deletion scope.
11.5 Cross-claim netting
The original lifecycle theorem applies to one claim trace. Section 5.2 establishes attribution conservation under its joint-commit and evidence premises. It separates attribution from current funding capacity. General netting additionally requires the joint obligation transformer, exposure maps, authority, and legal basis in Section 5.1. Each netting construction must prove those family-specific requirements. Attribution conservation alone establishes neither novation, close-out netting, nor legal extinction of an obligation.
11.6 Family migration
Section 7 establishes preservation under authorized semantic carriage, including splits, coalescence, and delayed outcomes. Novation, restructuring, and conversion may change economic rights. Their family-specific transformations need authority, explicit lineage, and their own preservation laws. The carriage theorem preserves their recorded consequences without claiming that these different transformations are lossless changes of meaning.
11.7 Liveness
Section 3.6 establishes progress after finite sufficient cure under explicit availability, stability, and fairness premises. A provider action, tribunal decision, or discretionary judgment retains its named external owner. The engine preserves awaiting obligations while that input remains unresolved. A changing current-source head can require reevaluation. The remote protocol’s recovery and timing assumptions determine progress there.
12 Conclusion
The unpaid remainder of a claim depends on its rights and payment history. The family interface exposes those obligations while retaining the family’s own state and transition rules. The engine binds each decision to its request, semantics, times, authority status, and evidence. Under the stated laws, accepted traces preserve the family invariant and consume each occurrence at most once within the claim. Refusals and awaiting decisions extend the recorded history while preserving the private state.
Shared allocation makes payment attribution explicit across claims. The duty journal retains that attribution when positions split, coalesce, or move to another program. For resolved duties, the interval construction preserves ordered payments and returns without enumerating every quantum. Amendment and other economic changes retain their own authority and recorded lineage.
Historical replay requires the original content and complete retained inputs. Joint execution requires the stated serialization and recovery conditions. A later external use requires authority at its named stage. The resulting records support an exact account of the engine’s decisions; their legal effect and provider evidence remain governed by the admitted institutional rules.
References
[1] ACTUS Financial Research Foundation. ACTUS: The Algorithmic Representation of Financial Contracts. Technical Specification v1.1-843f7a3-2020-06-08, 2020. https://www.actusfrf.org/techspecs.
[2] R. G. Brown, J. Carlyle, I. Grigg, and M. Hearn. Corda: An Introduction. R3, August 2016. https://docs.r3.com/en/pdf/corda-introductory-whitepaper.pdf.
[3] A. Bernauer, S. Faro, R. Hämmerle, et al. Daml: A smart contract language for securely automating real-world multi-party business workflows. arXiv:2303.03749v1, 2023.
[4] Hyperledger Fabric. Read-Write Set Semantics. Current public documentation, accessed September 2026. https://hyperledger-fabric.readthedocs.io/en/latest/readwrite.html.
[5] Board of Governors of the Federal Reserve System. Regulation J: Collection of Checks and Other Items by Federal Reserve Banks and Funds Transfers Through Fedwire. 12 CFR Part 210, section 210.31 and Appendix A. Federal Reserve Regulation J collection page.
[6] C. D. Clack, V. A. Bakshi, and L. Braine. Smart contract templates: Foundations, design landscape and research directions. arXiv:1608.00771, 2016.
[7] United Kingdom. Companies Act 2006, sections 112 and 127. https://www.legislation.gov.uk/ukpga/2006/46/contents.
[8] A. Biryukov, D. Khovratovich, and S. Tikhomirov. Findel: Secure derivative contracts for Ethereum. In Financial Cryptography Workshops, 2017.
[9] International Swaps and Derivatives Association. Financial products Markup Language, Version 5.13 Recommendation. 12 May 2025. https://www.fpml.org/latest_news/isda-publishes-the-recommendation-for-fpml-version-5-13/.
[10] International Swaps and Derivatives Association. ISDA Common Domain Model Version 1.0: Design Definition Document. October 2017. https://www.isda.org/isda-solutions-infohub/cdm/.
[11] FINOS. Common Domain Model. Current repository and documentation, accessed September 2026. https://github.com/finos/common-domain-model.
[12] P. Helland. Idempotence is not a medical condition. ACM Queue, 10(4), 2012. https://doi.org/10.1145/2160718.2160734.
[13] M. P. Herlihy and J. M. Wing. Linearizability: A correctness condition for concurrent objects. ACM Transactions on Programming Languages and Systems, 12(3):463–492, 1990. https://doi.org/10.1145/78969.78972.
[14] ISO 20022. Payments Clearing and Settlement: Payment Status Report (pacs.002). Message Definition Report, version 1.1, 2017. https://www.iso20022.org/sites/default/files/documents/D7/ISO20022_RTPG_pacs00200108_July_2017_v1_1.pdf.
[15] United Kingdom. Insolvency Act 1986, sections 238–241. https://www.legislation.gov.uk/ukpga/1986/45/contents.
[16] H. T. Kung and J. T. Robinson. On optimistic methods for concurrency control. ACM Transactions on Database Systems, 6(2):213–226, 1981. https://doi.org/10.1145/319566.319567.
[17] F. B. Schneider. Implementing fault-tolerant services using the state machine approach: A tutorial. ACM Computing Surveys, 22(4):299–319, 1990. https://doi.org/10.1145/98163.98167.
[18] United Kingdom. Law of Property Act 1925, section 136. https://www.legislation.gov.uk/ukpga/Geo5/15-16/20/section/136.
[19] LexiFi. Contract Description Language. Official technical note, accessed September 2026. https://www.lexifi.com/blog/structured-thoughts/contract-description-language/.
[20] R. Lorgat. How Compliance Composes. Companion manuscript, September 2026.
[21] R. Lorgat. The Claim as Primitive. Companion manuscript, September 2026.
[22] R. Lorgat. One Entity in Many Jurisdictions. Companion manuscript, September 2026.
[23] R. Lorgat. Op: A Typed Bytecode for Compliance-Carrying Operations. Companion manuscript, September 2026.
[24] R. Lorgat. Recourse. Companion manuscript, September 2026.
[25] R. Lorgat. AMM-Aware Central Counterparty Risk. Companion manuscript, September 2026.
[26] P. Lamela Seijas and S. Thompson. Marlowe: Financial contracts on blockchain. In Leveraging Applications of Formal Methods, 2018.
[27] G. C. Necula. Proof-carrying code. In Proceedings of POPL, pp. 106–119, 1997.
[28] S. Peyton Jones, J.-M. Eber, and J. Seward. Composing contracts: An adventure in financial engineering. In Proceedings of ICFP, pp. 280–292, 2000.
[29] Committee on Payment and Settlement Systems and International Organization of Securities Commissions. Principles for Financial Market Infrastructures. Bank for International Settlements, 2012. Principle 8, key considerations 1 and 3. https://www.bis.org/cpmi/publ/d101a.pdf.
[30] European Parliament and Council. Directive 98/26/EC on settlement finality in payment and securities settlement systems, Articles 3 and 5. https://eur-lex.europa.eu/eli/dir/1998/26/oj.
[31] R. E. Strom and S. Yemini. Typestate: A programming language concept for enhancing software reliability. IEEE Transactions on Software Engineering, 12(1):157–171, 1986.
[32] SWIFT. Cover Payments: Market Practice Guidelines. MPG FI CT (COV) number 12, p. 13, April 2026. https://www.swift.com/swift-resource/252248/download.
[33] World Wide Web Consortium. Bitstring Status List v1.0. W3C Recommendation, 15 May 2025. https://www.w3.org/TR/vc-bitstring-status-list/.