When Is a Transaction Complete?

A transaction is typically treated as complete when the price layer clears: cash is paid and goods/services are delivered. That completion rule is coherent when the only institutionally relevant ledger is the price ledger. But in capitalized economies—where a distinct capitalization layer continuously assigns and updates long-run value—transactions generate not only contemporaneous surplus but also updates to capitalized claims. Existing economic and empirical language often treat these updates as (i) expensed incentives (rebates, points), (ii) valuation statistics (CLV), or (iii) governance narratives (voice), while the platform pricing literature treats expensed transfers as the principal intertemporal lever in multi-sided markets. What is missing is the institutional primitive that makes long-run value part of transaction completion: an executable mapping from transaction facts to a recomputable, final, hard-budgeted, version-bound long-term settlement object. This paper formalizes two-layer transactions (TLT) and long-term layer settlement (LTL). LTL requires five institutional properties: (1) existence of a settlement mapping, (2) third-party recomputability, (3) verifiable finality (a testable non-rewrite boundary), (4) hard-budget discipline (excluding soft-budget promises), and (5) version binding (a cryptographic identifier binds mapping, accounting semantics, budget semantics, and finality parameters). We prove two theory results: a strict-inclusion theorem showing that under a hard budget, LTL strictly expands the implementable set relative to one-layer subsidy instruments when the long-run object is strictly more informative (Blackwell order); and a necessity-style impossibility showing that under state-dependent participation constraints and a hard budget, one-layer policies measurable only on contemporaneous objects cannot replicate LTL outcomes. Methodologically, we show that when finality/version/budget semantics are missing, many estimands are not “weakly identified”—they are undefined, because the treatment object itself drifts and treatment timing is not measurable in the analyst’s information set. We operationalize admissibility into an audit-grade standard, LAS v1.0, consisting of (i) a Minimum Recomputable Disclosure Row (MRDR Schema v1.0), (ii) an Institutional Integrity Score (IIS Rubric v1.0) with explicit handling of non-applicable dimensions, (iii) deterministic failure-window labeling, and (iv) a recompute pack that outputs verified MRDR, IIS, failure windows, and treatment panels with cryptographic digests. We provide three evidence modules with explicit hierarchy: (1) MR-CF (causal): a minimal recomputable causal fact—an RR-gated continuous-exposure event study on version-pure, finalized transactions, producing an auditable manifest and pass/fail rule card; (2) MRF (descriptive): a minimal reproducible descriptive fact using public data around a known program revision date (replication only; no behavioral causality claim); (3) Institutional witnesses: (i) an on-chain timelock template showing recomputable queue→execute finality, and (ii) an off-chain template showing that blockchains are sufficient but not necessary for LAS-admissible finality/version binding.