A letter to shareholders, clients, operators, and the people entrusted with market continuity
For most of the history of ➔ That separation became nor ➔ Databricks describes a dif ➔ For an institutional tradi ➔ I approach this subject fr
For most of the history of electronic trading, we accepted a division that was useful for engineers but costly for institutions. Online transaction processing systems recorded the immediate state of the business. Online analytical processing systems explained the business later. OLTP accepted orders, amendments, cancellations, allocations, approvals, and operational state changes. OLAP calculated exposure, liquidity, profitability, conduct indicators, and historical patterns. Between them stood queues, change-data-capture systems, replication jobs, caches, serving databases, semantic layers, and reconciliation controls.
That separation became normal, but normal is not the same as inevitable. It created an institutional tax. The transaction system knew what had just happened but lacked broad history. The analytical system understood history but was behind the market. The application serving layer was fast but often held another copy of the truth. Every copy introduced a delay, a control boundary, an ownership dispute, and an opportunity for the same position to acquire several meanings.
Databricks describes a different direction as Lake Transactional Analytical Processing, or LTAP. The objective is not to force every workload into one engine. The objective is to make transactional state, streaming computation, analytical history, application serving, machine learning, and governance parts of one coherent data system. Lakebase provides a fully managed PostgreSQL-compatible transactional database integrated with the Databricks platform. Structured Streaming Real-Time Mode reduces the latency assumptions inherited from micro-batch processing. Lakehouse Real-Time, also written Lakehouse//RT, provides low-latency, high-concurrency SQL serving directly over governed Delta Lake and Apache Iceberg tables. Unity Catalog supplies the common permissions, lineage, discovery, and policy boundary.
For an institutional trading desk, this is not an architectural fashion. It is an attempt to remove the gap between market reality and institutional awareness. A liquidity provider loses money not only when its forecast is wrong. It loses money when the order state, inventory state, hedge state, credit state, settlement state, or market-depth state is correct in one system and stale in another. A platform that reduces that gap can improve capital efficiency. A platform that merely hides the gap behind a fast interface can increase loss velocity.
I approach this subject from the background of a military combat engineer who became an institutional-grade trader providing market-liquidity-depth analysis to international trading desks. Combat engineering teaches that mobility is created through disciplined preparation. A bridge is not safe because it looks complete. It is safe because load, terrain, materials, redundancy, sequence, responsibility, and failure modes were made explicit. Trading infrastructure deserves the same standard. A ten-millisecond query is not useful if its data has been stale for ten minutes. A resilient database is not resilient if the application cannot identify authoritative state after failover. A unified catalog is not governance if service principals retain permissions nobody can explain.
Our task is therefore larger than deploying PostgreSQL beside a lakehouse. We must build an operating system for market truth. It must accept high-frequency transactional change, process events continuously, serve governed analytical context at application latency, and preserve enough evidence for risk, compliance, finance, operations, and shareholders to understand what occurred. It must be fast enough for the decision and careful enough for the consequence.