Membership & Participation

Every major financial institution made the same decision that early digital photographers made when storage was expensive: compress the raw signal into a manageable representation and discard the rest. For photographers, that meant JPEG. For financial institutions, it meant posting transactions to pre-aggregated balance structures — a deliberate, rational, economically justified decision that permanently discarded the attributes that did not fit the pre-defined posting key.

This collection of nine papers is the formal account of why the uncompressed architecture is correct, how it is built, and what it makes possible — from the theoretical foundation of the minimum cost point through the AI consequence, the production existence proofs, the instrument profitability and temporal perspective capabilities, the control and migration architecture, and the inter-enterprise frontier where the same logic eliminates the reconciliation tax between counterparties entirely. It is written for the practitioners, architects, and technologists who will do the building.

Nine papers in three volumes: Volume I — Theoretical Foundations (Papers 1–2) · Volume II — The Core Architecture (Papers 3–7) · Volume III — Transition and Extension (Papers 8–9)

Modern enterprise financial systems maintain multiple redundant balance structures — source system product ledgers, subledgers, general ledgers, data warehouse dimensions, and analytical datamarts — rooted in an architectural tradeoff made when computing was expensive: aggregate transactions into balances at posting time to avoid reprocessing history at query time. This paper argues that this tradeoff defines a formal mathematical cost curve with an interior minimum cost point, that virtually all major enterprises have drifted far to the right of that point, and that the cost of that drift is measurable in two compounding dimensions: (1) direct storage and reconciliation overhead from unnecessary balance duplication, and (2) the structural inability to answer queries that balance hierarchies cannot satisfy because the posting projection permanently discards the attributes they require.

Two production natural experiments at enterprise scale — a major U.S. property & casualty insurer's 25.5-billion-record event repository and a Tier-1 global bank Instrument Ledger that reduced a projected 46-terabyte analytical footprint to 250 gigabytes (a 99.5% reduction) — demonstrate that the minimum cost architecture is operationally feasible and economically decisive.

Keywords: REA framework · minimum cost curve · financial data architecture · event-sourced ledger · lossy compression · posting process · GenevaERS · instrument ledger · reconciliation cost · balances as inventory

The dominant paradigm for enterprise AI training on structured business data relies on tabular extracts from balance-based systems — product subledgers, General Ledgers, data warehouses, and OLAP cubes — that represent the accumulated output of decades of transaction posting processes. This paradigm systematically hobbles enterprise machine learning because transaction posting constitutes lossy compression in the Shannon sense. Posting destroys the multi-dimensional structure of business reality: it strips out counterparty context, severs causal event linkages, and collapses complex temporal processes into artificial "true-up" balance adjustments. When machine learning models are trained on these compressed balance marts, they are trapped in a "flat painting" — unable to discover latent causal patterns, anticipate non-linear outcomes across any business domain, or generalize beyond the fixed dimensional projections that were anticipated at posting time.

This paper introduces the Pre-Hierarchy Event Repository — an enterprise architecture in which raw economic business events (Resource-Event-Agent primitives) are preserved at full resolution as the primary book of record, with positions derived at query time — and establishes the information-theoretic basis for why pre-hierarchy corpora provide a higher information ceiling for Transformer-based sequence models and tabular Foundation Models. An empirical experimental design is proposed using a major U.S. property & casualty insurer's 25.5-billion-record event corpus spanning 25 years to quantitatively measure the information gain of pre-hierarchy sequence modeling over conventional balance-derived features.

Keywords: Enterprise AI · Foundation Models · Tabular Data · Event-Driven Architecture · Information Theory · Lossy Compression · REA Ontology · Pre-Hierarchy Training Corpora · Machine-Readable Contracts · GenevaERS